How artificial intelligence is used in scientific research, on the record.https://aixsci.org/https://aixsci.org/articles/aix-00217/https://aixsci.org/articles/aix-00217/Astronomers built BRIDGE, a model of how matter clumps into voids, sheets, filaments and knots, and fitted it to mock galaxy counts by gradient-based sampling to recover the universe's initial conditions.Fri, 09 Oct 2026 11:41:15 GMThttps://aixsci.org/articles/aix-00218/https://aixsci.org/articles/aix-00218/Researchers built Pheasy, a code that extracts the forces binding atoms in a crystal by fitting regularised regression models to simulated force data, then used them to compute vibration spectra and heat conduction for silicon, tungsten disulfide and strontium titanate.Fri, 09 Oct 2026 11:40:58 GMThttps://aixsci.org/articles/aix-00219/https://aixsci.org/articles/aix-00219/Researchers estimated how much magnetic energy flows up through the Sun's visible surface in quiet regions. A convolutional neural network, DeepVel, trained on simulations, supplied the sideways velocities that telescopes cannot measure directly.Fri, 09 Oct 2026 11:40:50 GMThttps://aixsci.org/articles/aix-00220/https://aixsci.org/articles/aix-00220/Researchers rolled a wedge-shaped titanium alloy sheet to build in a range of strains, then annealed it and measured how its alpha phase changed shape. A neural network was fitted to those measurements to predict the change across process settings.Fri, 09 Oct 2026 11:40:46 GMThttps://aixsci.org/articles/aix-00221/https://aixsci.org/articles/aix-00221/Researchers trained a neural network on brightness measurements from the Kepler space telescope to work out how fast stars spin. The model produced a catalogue of rotation periods, each with a calibrated uncertainty range.Fri, 09 Oct 2026 11:40:42 GMThttps://aixsci.org/articles/aix-00222/https://aixsci.org/articles/aix-00222/Researchers trained small neural networks on quantum-chemistry calculations for pairs of perylene and tetracene molecules, then used them to build and solve the equations describing light absorption in clusters of up to fifty molecules.Fri, 09 Oct 2026 11:40:38 GMThttps://aixsci.org/articles/aix-00223/https://aixsci.org/articles/aix-00223/Researchers trained convolutional neural networks on simulated eROSITA X-ray images of 3285 galaxy clusters to predict cluster mass, then used saliency maps to see which pixels the networks relied on.Fri, 09 Oct 2026 11:40:35 GMThttps://aixsci.org/articles/aix-00227/https://aixsci.org/articles/aix-00227/Astronomers trawled repeated infrared sky surveys for objects that shift position, using a random forest classifier to rank more than a million candidate tracks so that only the most promising ones reached a human eye.Fri, 09 Oct 2026 11:40:31 GMThttps://aixsci.org/articles/aix-00235/https://aixsci.org/articles/aix-00235/Astronomers trained a probabilistic random forest on spectroscopically classified stars and galaxies, then used it to assign a class and a probability to each of the roughly 130 million sources in a near-infrared survey of the Magellanic Clouds.Fri, 09 Oct 2026 11:40:27 GMThttps://aixsci.org/articles/aix-00238/https://aixsci.org/articles/aix-00238/Researchers trained six kinds of machine learning classifier to judge, reading by reading, whether a star's brightness dip marks a planet crossing its face. The models did the judging; the experiment varied where their training labels came from.Fri, 09 Oct 2026 11:40:24 GMThttps://aixsci.org/articles/aix-00216/https://aixsci.org/articles/aix-00216/Researchers trained two neural networks on quantum-mechanical calculations of lead titanate, then used them to run atom-by-atom simulations of the crystal's heating. The models supplied the forces and the dipole moments that every reported trajectory and spectrum rests on.Fri, 09 Oct 2026 11:40:20 GMThttps://aixsci.org/articles/aix-00215/https://aixsci.org/articles/aix-00215/Researchers built a search pipeline for colliding neutron stars in LIGO data. A convolutional neural network, trained on simulated signals injected into real detector noise, produced the score used to decide which candidates count as detections.Fri, 09 Oct 2026 11:40:06 GMThttps://aixsci.org/articles/aix-00213/https://aixsci.org/articles/aix-00213/Researchers dissolved calcined magnesite ore in carbonated water while applying ultrasound, then trained small neural networks on the resulting measurements to predict the fraction of ultrasound energy converted in the reaction.Fri, 09 Oct 2026 11:40:00 GMThttps://aixsci.org/articles/aix-00245/https://aixsci.org/articles/aix-00245/Astronomers adapted image classifiers trained on a large sky survey to work on the much smaller image set from the International Liquid Mirror Telescope. The networks sorted genuine new objects from subtraction artefacts and grouped the real ones.Fri, 09 Oct 2026 11:39:55 GMThttps://aixsci.org/articles/aix-00248/https://aixsci.org/articles/aix-00248/Six off-the-shelf image classifiers were run on archival radio images of the disc around the star HD 142666. Their outputs, and the patterns they responded to, pointed the astronomers to a disturbance in the gas about 75 au from the centre.Fri, 09 Oct 2026 11:39:50 GMThttps://aixsci.org/articles/aix-00251/https://aixsci.org/articles/aix-00251/Researchers built a model of how mass is spread through the galaxy cluster Abell 2744, using archival telescope data. A neural network sorted 23 of the 225 cluster galaxies that went into the model.Fri, 09 Oct 2026 11:39:41 GMThttps://aixsci.org/articles/aix-00254/https://aixsci.org/articles/aix-00254/Researchers trained three deep-learning models to estimate the dispersion measure of a fast radio burst from its frequency–time image, using 180,000 simulated bursts, then tried the models on real CHIME/FRB detections.Fri, 09 Oct 2026 11:39:37 GMThttps://aixsci.org/articles/aix-00210/https://aixsci.org/articles/aix-00210/Researchers tested seven off-the-shelf language models on predicting the starting chemicals and heating temperatures for making solid materials, then used the models to write 28,548 synthetic recipes and trained a smaller model, SyntMTE, on them.Fri, 09 Oct 2026 11:39:32 GMThttps://aixsci.org/articles/aix-00209/https://aixsci.org/articles/aix-00209/Researchers trained a neural network on laboratory measurements of 115 lunar soil samples, including Chang'e-5 material, then used it to predict the abundances of six major oxides across the Moon from orbital camera data.Fri, 09 Oct 2026 11:39:26 GMThttps://aixsci.org/articles/aix-00208/https://aixsci.org/articles/aix-00208/Chemists trained small neural networks to predict how electric charge is shared between the atoms in a molecule, using only the pattern of bonds. The networks stood in for quantum-chemistry calculations that normally need atomic positions.Fri, 09 Oct 2026 11:38:51 GMThttps://aixsci.org/articles/aix-00206/https://aixsci.org/articles/aix-00206/Researchers trained a neural network on 673 published hardness measurements of metallic glasses, then used it in reverse to invent new alloy recipes. Five were cast and tested, and their measured hardness matched the model's predictions.Fri, 09 Oct 2026 11:38:42 GMThttps://aixsci.org/articles/aix-00205/https://aixsci.org/articles/aix-00205/Researchers rebuilt a standard statistical fitting method so it could handle molecules that bend into many shapes, then trained it on measured oxidation potentials and hydration energies. The model did all the property prediction.Fri, 09 Oct 2026 11:38:33 GMThttps://aixsci.org/articles/aix-00204/https://aixsci.org/articles/aix-00204/Astronomers trained neural networks to reproduce the output of the Meudon PDR code, a simulation of starlit interstellar gas that takes hours to run. The networks predicted the same spectral line brightnesses in a fraction of the time.Fri, 09 Oct 2026 11:38:22 GMThttps://aixsci.org/articles/aix-00195/https://aixsci.org/articles/aix-00195/A study examined a computer-predicted structure of the honey bee egg-yolk protein vitellogenin, made with the AlphaFold neural network, and proposed that its tail region swings over to cover the protein's fatty cargo pocket.Fri, 09 Oct 2026 11:38:14 GMThttps://aixsci.org/articles/aix-00159/https://aixsci.org/articles/aix-00159/Researchers probed the local atomic arrangement of two tin- and germanium-substituted bismuth vanadate oxide ion conductors. A Gaussian process model, trained on charges from quantum calculations, assigned charges across atomic models too large for those calculations to handle.Fri, 09 Oct 2026 08:49:42 GMThttps://aixsci.org/articles/aix-00158/https://aixsci.org/articles/aix-00158/Researchers compared AlphaFold-predicted shapes of known human BRICHOS proteins, distilled a shared core, and searched the predicted human proteome with it. The search picked out a little-studied protein called Out at First, or OAF.Fri, 09 Oct 2026 08:49:35 GMThttps://aixsci.org/articles/aix-00157/https://aixsci.org/articles/aix-00157/Researchers overexpressed 2577 E. coli proteins tagged with a fluorescent marker and imaged the cells. A supervised neural network sorted each cell's glow pattern into three categories, which set each protein's aggregation class.Fri, 09 Oct 2026 08:49:29 GMThttps://aixsci.org/articles/aix-00156/https://aixsci.org/articles/aix-00156/Researchers built a tool that reads a protein's amino acid sequence and marks which positions bind ATP. A pre-trained protein language model encoded each position, and two learned classifiers, combined by weighted sum, made the calls.Fri, 09 Oct 2026 08:49:19 GMThttps://aixsci.org/articles/aix-00155/https://aixsci.org/articles/aix-00155/Researchers trained a graph-based neural network, FFiTrNet, on 12,072 compounds to predict how densely a molecule packs in its crystal. The model produced every density figure reported for the test compounds, in place of measurement or quantum-chemical calculation.Fri, 09 Oct 2026 08:49:13 GMThttps://aixsci.org/articles/aix-00154/https://aixsci.org/articles/aix-00154/Researchers built a database of the stretches of human proteins that pass through cell membranes. They took predicted three-dimensional structures from AlphaFold and used a physics-based program to settle each model into a membrane.Fri, 09 Oct 2026 08:49:06 GMThttps://aixsci.org/articles/aix-00153/https://aixsci.org/articles/aix-00153/Researchers trained two neural networks, PepSeP1 and PepSeP6, to choose the amino acids of six-residue peptide fragments lodged in a protein's binding site. The networks produced the sequences; Rosetta software then refined and scored them.Fri, 09 Oct 2026 08:48:58 GMThttps://aixsci.org/articles/aix-00152/https://aixsci.org/articles/aix-00152/Serpins are proteins that disable enzymes, and the short loop that does the work is hard to locate from sequence alone. Researchers trained neural networks on expert-labelled serpins to mark that loop residue by residue.Fri, 09 Oct 2026 08:48:52 GMThttps://aixsci.org/articles/aix-00151/https://aixsci.org/articles/aix-00151/Researchers folded curated pairs of human proteins with AlphaFold-Multimer, then trained a random forest classifier, SPOC, to score which predicted pairings look real. SPOC was applied to 40,459 predictions among 286 genome maintenance proteins.Fri, 09 Oct 2026 08:48:47 GMThttps://aixsci.org/articles/aix-00150/https://aixsci.org/articles/aix-00150/Researchers trained a neural network on the three-dimensional shapes of salen-type metal complexes to judge whether each one acts as a single-molecule magnet, then ran it over about 20,000 structures from a crystal database.Fri, 09 Oct 2026 08:48:39 GMThttps://aixsci.org/articles/aix-00149/https://aixsci.org/articles/aix-00149/Researchers fed the hepatitis E virus replicase sequence to AlphaFold2, the machine-learning structure prediction tool, and used the predicted shapes to mark out five protein domains and locate where their substrates and metal ions sit.Fri, 09 Oct 2026 08:48:34 GMThttps://aixsci.org/articles/aix-00147/https://aixsci.org/articles/aix-00147/Researchers built a large database of calculated energies for perovskite oxides and used it to compare two families of graph neural network. The networks were the object of study: how well each could rank different arrangements of the same atoms.Fri, 09 Oct 2026 08:48:26 GMThttps://aixsci.org/articles/aix-00160/https://aixsci.org/articles/aix-00160/Researchers trained graph neural networks to predict two carbon dioxide uptake properties of aluminium-substituted zeolites straight from their atomic structure, replacing slow Monte Carlo simulations and then steering a search for structures hitting chosen uptake targets.Fri, 09 Oct 2026 08:48:20 GMThttps://aixsci.org/articles/aix-00161/https://aixsci.org/articles/aix-00161/Researchers built ASCOT, an online tool that constructs spherical nanoparticles atom by atom, relaxes them with classical physics and measures their structure. No machine learning was used; the measurements are meant as inputs for later models.Fri, 09 Oct 2026 08:48:11 GMThttps://aixsci.org/articles/aix-00162/https://aixsci.org/articles/aix-00162/Researchers built a Python tool, CCfrag, that cuts a protein sequence into overlapping pieces and has AlphaFold predict each one. The program then merges the predictions into a position-by-position picture of the whole chain.Fri, 09 Oct 2026 08:48:07 GMThttps://aixsci.org/articles/aix-00163/https://aixsci.org/articles/aix-00163/Researchers built a pipeline that designed short peptides to gather at the edge of protein droplets inside cells. A neural network learned to predict simulation results, and an optimiser searched sequences through it.Fri, 09 Oct 2026 08:48:02 GMThttps://aixsci.org/articles/aix-00164/https://aixsci.org/articles/aix-00164/Researchers ran AlphaFold2, AlphaFold2-Multimer, AlphaFold3, Boltz-1 and Chai-1 over 509 known protein–peptide complexes, then fed the same models deliberately degraded inputs to see what their predictions actually depend on.Fri, 09 Oct 2026 08:47:58 GMThttps://aixsci.org/articles/aix-00165/https://aixsci.org/articles/aix-00165/Researchers built flash memory cells with channels of molybdenum disulfide just a few nanometres thick and measured how they stored charge. A reinforcement-learning agent then tuned a compact circuit model until its curves matched the measurements.Fri, 09 Oct 2026 08:47:54 GMThttps://aixsci.org/articles/aix-00166/https://aixsci.org/articles/aix-00166/Researchers built CATHe, a small neural network that reads numerical summaries of protein sequences produced by a language model and sorts the sequences into CATH structural superfamilies. It annotated 4.62 million previously unassigned domains.Fri, 09 Oct 2026 08:47:51 GMThttps://aixsci.org/articles/aix-00167/https://aixsci.org/articles/aix-00167/Researchers trained a YOLOv3 object detector to spot frontal waves in faint airglow pictures from a weather satellite, then used its detections across eleven years to chart where and when the waves appear.Fri, 09 Oct 2026 08:47:46 GMThttps://aixsci.org/articles/aix-00168/https://aixsci.org/articles/aix-00168/Researchers built MICA, a method that feeds a cryo-electron microscopy density map and AlphaFold3's predicted chain structures into one trained network, which labels each point in the map before the atomic model is assembled.Fri, 09 Oct 2026 08:47:42 GMThttps://aixsci.org/articles/aix-00169/https://aixsci.org/articles/aix-00169/Researchers asked a structure-prediction program to model a hub protein alongside short stretches of its partners, then used the program's own confidence numbers to sort binders from non-binders and to flag candidate sites in 24 proteins.Fri, 09 Oct 2026 08:47:38 GMThttps://aixsci.org/articles/aix-00170/https://aixsci.org/articles/aix-00170/Researchers moulded 27 batches of a boron nitride-reinforced bioplastic and measured their mechanical properties. Five machine learning regression models were then fitted to the same data to predict strength, stiffness and hardness from the moulding settings.Fri, 09 Oct 2026 08:47:34 GMThttps://aixsci.org/articles/aix-00171/https://aixsci.org/articles/aix-00171/Researchers trained a model called CrysFormer to produce protein electron-density maps from two inputs: a Patterson map computed from diffraction data, and an incomplete predicted template with several residues missing.Fri, 09 Oct 2026 08:47:29 GMThttps://aixsci.org/articles/aix-00172/https://aixsci.org/articles/aix-00172/Researchers trained random forest classifiers on sequence-derived biophysical predictions to tell ordered, disordered and 'ambiguous' protein residues apart. The models' scores and rules were the evidence used to compare rival definitions of protein order.Fri, 09 Oct 2026 08:47:25 GMThttps://aixsci.org/articles/aix-00146/https://aixsci.org/articles/aix-00146/Researchers trained a simple statistical model on measured superconductors, then used it to estimate the critical temperature of about 153,000 known compounds from chemical composition alone. Sixty-four stable candidates were predicted above 250 K.Fri, 09 Oct 2026 08:44:55 GMThttps://aixsci.org/articles/aix-00145/https://aixsci.org/articles/aix-00145/Researchers built a design pipeline in which AlphaFold2 and a solenoid-classifying network scored randomly generated repeat sequences inside a genetic algorithm. Of 41 designs made in the laboratory, several alpha-solenoids behaved as modelled, and one was solved by X-ray crystallography.Fri, 09 Oct 2026 08:44:50 GMThttps://aixsci.org/articles/aix-00144/https://aixsci.org/articles/aix-00144/Researchers fitted two machine-learned interatomic potentials to quantum-mechanical calculations on the copper compound Cu7PS6, then used them in place of those calculations to simulate atomic structure and vibrations.Fri, 09 Oct 2026 08:44:46 GMThttps://aixsci.org/articles/aix-00143/https://aixsci.org/articles/aix-00143/Researchers built computer-predicted structures of the proteins PRMT5 and RUVBL1 in humans and two yeasts, and scored cancer mutations in them. AlphaFold 3 supplied the structures; a classifier called CHASM Plus rated each mutation as a likely driver or a bystander.Fri, 09 Oct 2026 08:44:40 GMThttps://aixsci.org/articles/aix-00142/https://aixsci.org/articles/aix-00142/Researchers trained a separate machine learning force field for each of seven quantum-mechanical descriptions of water, then used them to simulate liquid water and work out its structure, entropy, viscosity and diffusion.Fri, 09 Oct 2026 08:44:36 GMThttps://aixsci.org/articles/aix-00140/https://aixsci.org/articles/aix-00140/Researchers built M3Site, a model that reads a protein's sequence, its predicted three-dimensional shape and a written description of its function, then labels each building block as an active site of one of six functional kinds.Fri, 09 Oct 2026 08:44:24 GMThttps://aixsci.org/articles/aix-00139/https://aixsci.org/articles/aix-00139/Researchers built a loop that heats patches of thin film with a laser, reads the resulting crystal structure by X-ray, and lets software decide the next condition to try. Machine learning identified the phases and chose the experiments.Fri, 09 Oct 2026 08:44:10 GMThttps://aixsci.org/articles/aix-00138/https://aixsci.org/articles/aix-00138/Researchers assembled thousands of microbial proteins with no close known relatives, then used a neural network and structure comparison to mark which pairs probably share a job. Thirteen annotation tools were scored against those labels.Fri, 09 Oct 2026 08:44:01 GMThttps://aixsci.org/articles/aix-00136/https://aixsci.org/articles/aix-00136/Researchers trained a generative image model on nine years of solar observations, labelled by flare strength, and then used its invented images of the Sun as extra training material for flare classifiers and predictors.Fri, 09 Oct 2026 08:43:57 GMThttps://aixsci.org/articles/aix-00173/https://aixsci.org/articles/aix-00173/Researchers gathered fibre measurements from published electrospinning studies and trained seven machine-learning models to predict fibre diameter from six process settings. The models supply the predictions, served through a web app, which were then compared with freshly spun fibres.Fri, 09 Oct 2026 08:43:53 GMThttps://aixsci.org/articles/aix-00174/https://aixsci.org/articles/aix-00174/With no laboratory structure of the poppy enzyme REPI available, researchers built their model on an AlphaFold2 prediction, then ran conventional molecular simulations to watch how its two halves move and how the chemical passes between them.Fri, 09 Oct 2026 08:43:46 GMThttps://aixsci.org/articles/aix-00176/https://aixsci.org/articles/aix-00176/Researchers built fast neural-network models of the forces between hydrogen molecules and copper surfaces, growing the training data by letting committees of models flag the configurations they disagreed about and labelling those with quantum calculations.Fri, 09 Oct 2026 08:43:34 GMThttps://aixsci.org/articles/aix-00178/https://aixsci.org/articles/aix-00178/Researchers tested 255 new catalyst compositions for turning carbon dioxide into carbon monoxide. A machine learning model trained on measured activity chose which mixtures to make, and the lab results were fed back to retrain it.Fri, 09 Oct 2026 08:43:28 GMThttps://aixsci.org/articles/aix-00135/https://aixsci.org/articles/aix-00135/Astronomers used an unsupervised Gaussian Mixture Model on Gaia DR3 motions and distances to separate the members of thirteen open clusters from unrelated field stars, then measured how well the model performed.Fri, 09 Oct 2026 08:43:23 GMThttps://aixsci.org/articles/aix-00134/https://aixsci.org/articles/aix-00134/Researchers trained a neural network to stand in for a slow theoretical calculation of how cosmic gravity distorts galaxy images, using a meta-learning method so one network could be adapted to a new survey's galaxy distribution from 100 examples.Fri, 09 Oct 2026 08:43:18 GMThttps://aixsci.org/articles/aix-00133/https://aixsci.org/articles/aix-00133/Researchers fitted a range of machine learning regression models to existing emission measurements of aluminium and aluminium–graphene composites, then used the fitted models to produce emission curves for higher graphene contents that had not been measured.Fri, 09 Oct 2026 08:43:14 GMThttps://aixsci.org/articles/aix-00132/https://aixsci.org/articles/aix-00132/The Euclid Morphology Challenge tested five pieces of software on about 1.5 million simulated galaxies. One of the five measured galaxy shapes with neural networks, and a generative model produced one set of the test images.Fri, 09 Oct 2026 08:43:08 GMThttps://aixsci.org/articles/aix-00131/https://aixsci.org/articles/aix-00131/Researchers trained a neural network to stand in for costly quantum-mechanical calculations of chromium sulfides, then used it to search for stable atomic arrangements and to simulate a strained slab peeling into layers.Fri, 09 Oct 2026 08:43:03 GMThttps://aixsci.org/articles/aix-00130/https://aixsci.org/articles/aix-00130/Researchers wanted the free energy barrier for a single step in CO hydrogenation on a rhodium surface. A machine-learned model of the atomic forces stood in for quantum chemistry, making the 16 nanoseconds of biased molecular dynamics tractable.Fri, 09 Oct 2026 08:42:58 GMThttps://aixsci.org/articles/aix-00179/https://aixsci.org/articles/aix-00179/Researchers built MCGLPPI, a framework that turns protein complexes into coarse-grained graphs. Graph neural networks learned from these graphs to predict binding strength and to tell real biological interfaces from artefacts of crystal packing.Fri, 09 Oct 2026 08:42:51 GMThttps://aixsci.org/articles/aix-00180/https://aixsci.org/articles/aix-00180/Researchers built a training strategy, FSFP, that adapts large protein language models using only tens of measured mutants. The models then ranked mutants across a public benchmark and picked candidates for a DNA-copying enzyme tested in the lab.Fri, 09 Oct 2026 08:42:46 GMThttps://aixsci.org/articles/aix-00181/https://aixsci.org/articles/aix-00181/Researchers built LMSuccSite, a tool that predicts which lysines in a protein carry a succinyl tag, using only the protein's sequence. Numerical descriptions from a pre-trained protein language model replaced hand-designed sequence features.Fri, 09 Oct 2026 08:42:39 GMThttps://aixsci.org/articles/aix-00183/https://aixsci.org/articles/aix-00183/Researchers trained a convolutional neural network on simulated nanostructures to predict how each one bends light, then used its predictions to design and build a metasurface that collimates red and near-infrared laser beams.Fri, 09 Oct 2026 08:42:34 GMThttps://aixsci.org/articles/aix-00184/https://aixsci.org/articles/aix-00184/Researchers built MetaDegron, two trained models that read short protein sequences and score whether they are degrons, the tags that mark a protein for disposal. One model learns from computed structural features, the other from sequence alone.Fri, 09 Oct 2026 08:42:30 GMThttps://aixsci.org/articles/aix-00185/https://aixsci.org/articles/aix-00185/Researchers built PCBert-Kla, a tool that predicts which lysine residues in a protein carry a lactyl tag. A pretrained protein language model, ProtBert, supplied the sequence representation and was fine-tuned alongside the classifier.Fri, 09 Oct 2026 08:42:27 GMThttps://aixsci.org/articles/aix-00186/https://aixsci.org/articles/aix-00186/Researchers fine-tuned ProtBERT, a model pretrained on protein sequences, to predict three physical properties of intrinsically disordered proteins straight from their amino acid sequence, using values first computed by molecular simulation.Fri, 09 Oct 2026 08:42:23 GMThttps://aixsci.org/articles/aix-00187/https://aixsci.org/articles/aix-00187/Researchers applied FFLUX, a force field whose atomic energies and electrical descriptions come from Gaussian process regression models, to the ambient and high-pressure crystal forms of formamide, then compared the predicted structures, vibrations and spectra with experiment and standard quantum calculations.Fri, 09 Oct 2026 08:42:19 GMThttps://aixsci.org/articles/aix-00188/https://aixsci.org/articles/aix-00188/Researchers turned 1111 antibody heavy chain sequences into numerical descriptors and trained tree-based classifiers, an ensemble and a Transformer to predict which of five antigens each antibody targets, in place of laboratory binding tests.Fri, 09 Oct 2026 08:42:15 GMThttps://aixsci.org/articles/aix-00189/https://aixsci.org/articles/aix-00189/Researchers tested six deep-learning tools on G protein-coupled receptors: three predicted how known peptides sit in 113 receptor complexes, three designed new peptides for three receptors. Every result came from the software itself.Fri, 09 Oct 2026 08:42:11 GMThttps://aixsci.org/articles/aix-00190/https://aixsci.org/articles/aix-00190/Researchers built MVGFormer, a neural network that labels the contents of three-dimensional cryo-electron tomography volumes. The model does the labelling itself, reading each volume from three perpendicular views and picking out particles that people would otherwise mark by hand.Fri, 09 Oct 2026 08:42:07 GMThttps://aixsci.org/articles/aix-00191/https://aixsci.org/articles/aix-00191/Astronomers built LRPayne, a method for pulling temperatures, gravities and 24 elemental abundances out of low-resolution starlight. A neural network trained on 70,000 simulated spectra replaces the physics code inside the fitting loop.Fri, 09 Oct 2026 08:42:03 GMThttps://aixsci.org/articles/aix-00192/https://aixsci.org/articles/aix-00192/Researchers ran three released machine-learning tools for tidying up cryo-electron microscopy density maps across a panel of deposited structures, then compared the results with conventional methods. The AI tools were the objects under test rather than the researchers' own creations.Fri, 09 Oct 2026 08:41:59 GMThttps://aixsci.org/articles/aix-00193/https://aixsci.org/articles/aix-00193/A team modelled the stretching and heating of nanotubes rolled from DHQ, a carbon sheet, using classical molecular dynamics. The forces between atoms came from a model trained on quantum-mechanical calculations, because standard empirical force fields did not reproduce the material.Fri, 09 Oct 2026 08:41:55 GMThttps://aixsci.org/articles/aix-00194/https://aixsci.org/articles/aix-00194/Researchers built AMDEN, a diffusion model that writes out the atomic arrangements of disordered materials to match properties asked for in advance, such as stiffness or lithium content, in place of searching by simulation.Fri, 09 Oct 2026 08:41:50 GMThttps://aixsci.org/articles/aix-00129/https://aixsci.org/articles/aix-00129/A thesis built machine learning tools for supernova cosmology using only brightness measurements. Convolutional networks classified supernova types and predicted redshift distributions, standing in for the spectra and expert judgement normally needed.Fri, 09 Oct 2026 08:41:42 GMThttps://aixsci.org/articles/aix-00128/https://aixsci.org/articles/aix-00128/Researchers turned viral protein sequences into colour-coded images and fine-tuned three off-the-shelf image-recognition networks to say which proteins form part of a virus's physical shell, then measured how confident those networks were.Thu, 08 Oct 2026 22:08:39 GMThttps://aixsci.org/articles/aix-00126/https://aixsci.org/articles/aix-00126/Researchers swapped the starting guess inside a structure-search method for a pretrained machine learning model of atomic forces, then tested how often the search found the lowest-energy arrangement of silica, a copper cluster and a titanium dioxide surface.Thu, 08 Oct 2026 22:08:33 GMThttps://aixsci.org/articles/aix-00125/https://aixsci.org/articles/aix-00125/Researchers simulated Euclid-like images of stars in one broad colour band and trained machine learning models to sort each star into one of 13 spectral types. The classifiers did the sorting; a support vector machine worked from how well modelled blur patterns matched each image.Thu, 08 Oct 2026 22:08:28 GMThttps://aixsci.org/articles/aix-00124/https://aixsci.org/articles/aix-00124/Researchers assembled six sets of retinal proteins, one healthy control and five tied to eye disease, then used off-the-shelf prediction tools to estimate how disordered each protein is and how readily it might form liquid droplets. No laboratory measurements were made.Thu, 08 Oct 2026 22:08:23 GMThttps://aixsci.org/articles/aix-00123/https://aixsci.org/articles/aix-00123/Researchers trained two networks on a library of simulated black hole images so that physical settings can be read off an image, and an image produced from settings. The networks stood in for the slower ray-tracing simulation.Thu, 08 Oct 2026 22:08:17 GMThttps://aixsci.org/articles/aix-00122/https://aixsci.org/articles/aix-00122/Astronomers trained a Random Forest classifier on low-resolution Gaia spectra of nearby white dwarfs, using existing catalogue labels, then applied it to assign spectral types to 9,446 stars that had none.Thu, 08 Oct 2026 22:08:13 GMThttps://aixsci.org/articles/aix-00121/https://aixsci.org/articles/aix-00121/Researchers built a classifier that guesses which part of the mitochondrion a protein sits in, using only its amino acid sequence. A convolutional neural network did the sorting, with gradient boosting trimming the input features.Thu, 08 Oct 2026 22:08:04 GMThttps://aixsci.org/articles/aix-00119/https://aixsci.org/articles/aix-00119/Researchers trained a neural network to reproduce quantum-mechanical forces between atoms in calcium oxide, then used it to run molecular dynamics on supercells of 10,648 to 17,280 atoms and compute melting temperatures up to 20 GPa.Thu, 08 Oct 2026 22:07:07 GMThttps://aixsci.org/articles/aix-00118/https://aixsci.org/articles/aix-00118/Astronomers studied sequences of solar images from the GONG telescope network to find oscillations on the Sun. Two convolutional neural networks were trained to stand in for a Bayesian fitting step that otherwise took a minimum of ten seconds per pixel.Thu, 08 Oct 2026 22:07:02 GMThttps://aixsci.org/articles/aix-00117/https://aixsci.org/articles/aix-00117/Researchers built Periodic-TDL, a model that reads a polymer's repeating unit as a shape and predicts its properties. It forecast glass transition temperatures for an enumerated library of 48,208 polymers; three were then made and measured.Thu, 08 Oct 2026 22:06:56 GMThttps://aixsci.org/articles/aix-00116/https://aixsci.org/articles/aix-00116/Researchers built PNAbind, a graph neural network trained on protein surface shapes and chemistry, to say whether a protein binds DNA or RNA and which of its residues do the binding.Thu, 08 Oct 2026 22:06:52 GMThttps://aixsci.org/articles/aix-00115/https://aixsci.org/articles/aix-00115/Researchers assembled CARE, a set of curated enzyme datasets and held-out test splits, then ran machine-learning models on them, including CREEP, a model they built by fine-tuning existing language models of proteins and reactions.Thu, 08 Oct 2026 22:06:43 GMThttps://aixsci.org/articles/aix-00114/https://aixsci.org/articles/aix-00114/Researchers built cMBDF, a fixed-size numerical description of each atom's surroundings, and used it to train kernel ridge regression models that predict quantum properties of small molecules from their geometry alone.Thu, 08 Oct 2026 22:06:37 GMThttps://aixsci.org/articles/aix-00113/https://aixsci.org/articles/aix-00113/Researchers trained a neural network interatomic potential to reproduce made-up forces that appear when known structures are jostled. Random clumps of atoms were then relaxed on that learned surface to produce molecules and crystals.Thu, 08 Oct 2026 22:06:31 GMThttps://aixsci.org/articles/aix-00112/https://aixsci.org/articles/aix-00112/Researchers compared two ways of turning bacterial protein sequences into numbers for predicting antimicrobial resistance. One used hand-computed chemical descriptors; the other used frozen embeddings from the ESM-2 protein language model, fed to five classifiers.Thu, 08 Oct 2026 22:06:25 GMThttps://aixsci.org/articles/aix-00111/https://aixsci.org/articles/aix-00111/Researchers padded the symmetry description of inorganic crystals with zeros so a single variational autoencoder could handle compositions with different numbers of elements. The network learned to reconstruct and generate candidate structures; pretrained potentials then relaxed and scored them.Thu, 08 Oct 2026 22:06:20 GMThttps://aixsci.org/articles/aix-00110/https://aixsci.org/articles/aix-00110/Researchers stressed 51 antibodies and measured chemical damage at thousands of sites with an automated robotic workflow. A protein language model turned those measurements into a predictor of which sites degrade, and by how much, from sequence alone.Thu, 08 Oct 2026 22:06:14 GMThttps://aixsci.org/articles/aix-00108/https://aixsci.org/articles/aix-00108/Researchers built OxPot, an open dataset of oxidation potentials for 15,238 organic molecules, using a fitted linear relationship to convert quantum-chemical calculations into potentials, then trained several machine learning models on the result.Thu, 08 Oct 2026 22:06:00 GMThttps://aixsci.org/articles/aix-00107/https://aixsci.org/articles/aix-00107/Researchers built StackRPred, a classifier that labels a plant protein sequence as a resistance protein or not. Six machine learning models feed a seventh, which makes the final call from features derived from a residue energy matrix.Thu, 08 Oct 2026 22:05:54 GMThttps://aixsci.org/articles/aix-00106/https://aixsci.org/articles/aix-00106/Volunteers marked asteroid streaks in archival Hubble images, and those marks trained an object-detection model that was then run across the archive. The combined search yielded 1 701 trails, 670 of them linked to known Solar System objects.Thu, 08 Oct 2026 22:05:49 GMThttps://aixsci.org/articles/aix-00104/https://aixsci.org/articles/aix-00104/Astronomers used two neural networks to classify the shapes of about 400,000 galaxies in JWST's COSMOS-Web survey, and to spot stellar bars, then counted each shape across cosmic time.Thu, 08 Oct 2026 22:05:35 GMThttps://aixsci.org/articles/aix-00103/https://aixsci.org/articles/aix-00103/Researchers made chitosan nanoparticles using olive leaf extract across 50 planned experiments, then trained a small neural network on those runs to predict how much material each set of conditions would produce.Thu, 08 Oct 2026 22:05:31 GMThttps://aixsci.org/articles/aix-00101/https://aixsci.org/articles/aix-00101/Researchers used quantum-mechanical simulations to work out the structure, vibrations and heat-to-electricity behaviour of three sodium–bismuth compounds. A random forest and a small neural network were then trained on those simulated results to predict one compound's efficiency figure.Thu, 08 Oct 2026 22:05:25 GMThttps://aixsci.org/articles/aix-00100/https://aixsci.org/articles/aix-00100/Researchers compared off-the-shelf image-recognition networks, pre-trained on everyday photographs or on galaxy images, against a purpose-built network for deciding which nightly survey alerts are real transients. The models did the sifting and the outlier hunting themselves.Thu, 08 Oct 2026 22:05:17 GMThttps://aixsci.org/articles/aix-00099/https://aixsci.org/articles/aix-00099/Astronomers used the motions of millions of Gaia stars to weigh a patch of our Galaxy's disk. Two machine-learning products supplied the inputs: a smoothed map of how densely stars sit, and predicted line-of-sight velocities for stars missing them.Thu, 08 Oct 2026 22:05:11 GMThttps://aixsci.org/articles/aix-00097/https://aixsci.org/articles/aix-00097/Researchers trained two small neural networks to stand in for slow calculations of how loud a space-based gravitational wave signal would be, and used them to work out what a future survey of such signals could reveal.Thu, 08 Oct 2026 22:04:59 GMThttps://aixsci.org/articles/aix-00096/https://aixsci.org/articles/aix-00096/Researchers used two off-the-shelf deep learning models, ProteinMPNN and ESM-IF1, to rewrite the gripping parts of T cell receptors on fixed structural scaffolds, then compared the designs with a physics-based method and with the natural sequences.Thu, 08 Oct 2026 22:04:54 GMThttps://aixsci.org/articles/aix-00095/https://aixsci.org/articles/aix-00095/Researchers trained neural networks on simulated universes to work out how massive a galaxy's halo of dark matter is, using only its own mass and the positions of nearby galaxies. The networks produced the predictions the study's conclusions rest on.Thu, 08 Oct 2026 22:04:47 GMThttps://aixsci.org/articles/aix-00094/https://aixsci.org/articles/aix-00094/Researchers simulated carbon dioxide gas inside two porous crystalline materials to track how it is held and how it moves. A pre-trained neural network, ANI-2x, supplied the atomic energies and forces that drove every simulation.Thu, 08 Oct 2026 22:04:33 GMThttps://aixsci.org/articles/aix-00093/https://aixsci.org/articles/aix-00093/Researchers trained a neural network on the IllustrisTNG-300 simulation to predict the electron pressure field inside galaxy clusters directly from dark matter particles, standing in for a much costlier simulation of the gas physics.Thu, 08 Oct 2026 22:04:26 GMThttps://aixsci.org/articles/aix-00092/https://aixsci.org/articles/aix-00092/Researchers purified the bacterial proteins NorQ and NorD, imaged them by electron microscopy at low resolution, and used AlphaFold to predict their structures. The predictions were fitted into the blurry density, and mutations tested the arrangement they suggested.Thu, 08 Oct 2026 22:04:21 GMThttps://aixsci.org/articles/aix-00091/https://aixsci.org/articles/aix-00091/Researchers trained a neural network to predict the forces between atoms in the molecular ferroelectric HdabcoClO4, then used it to run molecular dynamics on crystals of 11,232 atoms at temperatures from 120 to 500 K.Thu, 08 Oct 2026 22:04:15 GMThttps://aixsci.org/articles/aix-00090/https://aixsci.org/articles/aix-00090/Astronomers built a grid of physical models of the dusty discs around young stars, then trained decision-tree regressors to run the models backwards, reading gas mass, dust mass and disc size from telescope measurements of 34 discs.Thu, 08 Oct 2026 22:04:08 GMThttps://aixsci.org/articles/aix-00089/https://aixsci.org/articles/aix-00089/Researchers built HERMES, a family of neural networks that read the atoms around a single site in a protein and rank how each of the 20 amino acids would sit there. Every stability, binding and antigen result reported is a model prediction.Thu, 08 Oct 2026 22:04:02 GMThttps://aixsci.org/articles/aix-00088/https://aixsci.org/articles/aix-00088/Researchers computed higher-level optical spectra for about 6,000 crystals, then retrained an existing graph neural network on them. The network predicted the spectra of unseen materials, standing in for calculations that took 193,639 CPU hours.Thu, 08 Oct 2026 21:50:34 GMThttps://aixsci.org/articles/aix-00087/https://aixsci.org/articles/aix-00087/Researchers applied a method called Deep Potential to a computer-simulated barred galaxy, using neural networks to learn the stars' motions and then fit the galaxy's gravitational field and the rotation speed of the bar.Thu, 08 Oct 2026 21:49:27 GMThttps://aixsci.org/articles/aix-00086/https://aixsci.org/articles/aix-00086/Researchers simulated fast radio bursts of six different shapes and used them to train convolutional neural networks to sort bursts into five shape classes. The networks were then tested on 535 real bursts from the first CHIME/FRB catalogue.Thu, 08 Oct 2026 21:49:22 GMThttps://aixsci.org/articles/aix-00084/https://aixsci.org/articles/aix-00084/Astronomers built a pipeline that learns to infer cosmological parameters from mock catalogues of X-ray galaxy clusters. Two trained networks did the work: one compressed each catalogue into a short summary, the other turned that summary into a probability distribution over parameters.Thu, 08 Oct 2026 21:49:06 GMThttps://aixsci.org/articles/aix-00083/https://aixsci.org/articles/aix-00083/Researchers trained a small generative model on simulations of floppy proteins, then used it to produce new backbone shapes. The model stood in for further simulation, turning out 50,000 conformations of one protein in about 50 seconds.Thu, 08 Oct 2026 21:49:01 GMThttps://aixsci.org/articles/aix-00082/https://aixsci.org/articles/aix-00082/Researchers trained U-Net neural networks on simulated galaxy clusters to convert maps of total mass into maps of the X-ray and microwave signals clusters give off, then applied them to dark-matter-only simulations.Thu, 08 Oct 2026 21:48:55 GMThttps://aixsci.org/articles/aix-00081/https://aixsci.org/articles/aix-00081/Researchers built StackGlyEmbed, a tool that predicts which points in a human protein carry an attached sugar chain. Three pre-trained protein language models turned each candidate site into numbers, and a stack of classifiers made the call.Thu, 08 Oct 2026 21:48:48 GMThttps://aixsci.org/articles/aix-00080/https://aixsci.org/articles/aix-00080/Researchers trained a classifier called BROWDIE on simulated and catalogued infrared brightnesses to pick out the coldest kinds of brown dwarf. Applied to a region of the UKIDSS survey, it returned 132 candidates, 118 T dwarfs and 14 Y dwarfs.Thu, 08 Oct 2026 21:48:42 GMThttps://aixsci.org/articles/aix-00079/https://aixsci.org/articles/aix-00079/Researchers trained a 3D U-Net to label every point in a medium-resolution cryo-electron-microscopy map as helix, beta-sheet or background, and compared five ways of penalising its mistakes during training. The combination of focal and Dice loss scored highest for beta-sheets.Thu, 08 Oct 2026 21:48:29 GMThttps://aixsci.org/articles/aix-00078/https://aixsci.org/articles/aix-00078/Researchers trained a machine learning model to imitate quantum chemistry calculations, then used it to simulate water squeezed between graphene sheets. The simulations tracked where a single positive or negative charge in the water prefers to sit.Thu, 08 Oct 2026 21:48:21 GMThttps://aixsci.org/articles/aix-00077/https://aixsci.org/articles/aix-00077/Researchers searched DESI DR1 quasar spectra for quasars bending the light of galaxies behind them. One neural network flagged 494 candidates; a second estimated the background galaxy's distance. Seven candidates were graded A after human inspection.Thu, 08 Oct 2026 21:48:15 GMThttps://aixsci.org/articles/aix-00076/https://aixsci.org/articles/aix-00076/Researchers trained a neural network called DeepClean on data from LIGO's third observing run to subtract 60 Hz electrical noise from gravitational-wave recordings, then checked the cleaned data against a search pipeline and a parameter-estimation run.Thu, 08 Oct 2026 21:48:06 GMThttps://aixsci.org/articles/aix-00075/https://aixsci.org/articles/aix-00075/Astronomers trained a neural network to judge, from a star's position and motion alone, whether it was born in the Milky Way or swallowed from a smaller galaxy. It flagged 160,146 such stars among more than 27 million.Thu, 08 Oct 2026 21:47:48 GMThttps://aixsci.org/articles/aix-00074/https://aixsci.org/articles/aix-00074/Researchers ran the human PIEZO1 and PIEZO2 protein sequences through a set of off-the-shelf sequence predictors, several of them neural networks, to estimate which stretches are disordered, which may bind partners, and how clinical mutations line up with them.Thu, 08 Oct 2026 21:47:39 GMThttps://aixsci.org/articles/aix-00073/https://aixsci.org/articles/aix-00073/Researchers trained two neural networks to read ion measurements from NASA's MAVEN spacecraft and label which plasma region it was flying through. The networks were then run over all available observations from 2014 to 2025.Thu, 08 Oct 2026 21:47:25 GMThttps://aixsci.org/articles/aix-00072/https://aixsci.org/articles/aix-00072/Researchers split protein sequences into short segments they call protein words, using the internal attention patterns of the ESM2 language model, then trained a second model to link those words to molecular functions.Thu, 08 Oct 2026 21:47:19 GMThttps://aixsci.org/articles/aix-00071/https://aixsci.org/articles/aix-00071/Researchers used quantum-chemistry calculations to work out whether lone dopant atoms in metal surfaces stay put or sink inwards, bare and with three small molecules attached. A neural network was then trained on those results to predict the same energies quickly.Thu, 08 Oct 2026 21:47:12 GMThttps://aixsci.org/articles/aix-00070/https://aixsci.org/articles/aix-00070/Researchers built a search that proposes internal structures for silicon-carbide-reinforced aluminium composites meeting set strength and toughness targets. A trained neural network predicted each candidate's properties in place of running a fresh simulation.Thu, 08 Oct 2026 21:47:07 GMThttps://aixsci.org/articles/aix-00069/https://aixsci.org/articles/aix-00069/Researchers used two neural networks to reconstruct how fast the universe has been expanding at different distances, then used that curve to work out distances to hundreds of gamma-ray bursts and fit a known relation between their energies.Thu, 08 Oct 2026 21:47:01 GMThttps://aixsci.org/articles/aix-00068/https://aixsci.org/articles/aix-00068/Researchers trained models to predict how fast six marine engineering steels corrode in seawater. A genetic algorithm tuned the models, and a mixture model plus a generative network produced synthetic extra training samples.Thu, 08 Oct 2026 21:46:47 GMThttps://aixsci.org/articles/aix-00066/https://aixsci.org/articles/aix-00066/Researchers assembled a database of over 25,000 computed materials, labelled some as promising or unpromising hosts for quantum devices, then trained four classifiers to sort the rest. All three labelling schemes and all four methods agreed on 47 candidates.Thu, 08 Oct 2026 21:46:39 GMThttps://aixsci.org/articles/aix-00065/https://aixsci.org/articles/aix-00065/Researchers built IECata, a neural network that reads an enzyme's amino acid sequence and a chemical's structure and predicts catalytic efficiency. The model also reports how confident it is, and highlights which residues and atoms it attended to.Thu, 08 Oct 2026 21:46:31 GMThttps://aixsci.org/articles/aix-00063/https://aixsci.org/articles/aix-00063/Researchers added a fitted correction for the unfolded state of a mutated protein to scores from two machine-learning predictors, ESM-IF1 and Pythia. The models supplied the folded-state scores; a small linear model learned the correction.Thu, 08 Oct 2026 21:45:46 GMThttps://aixsci.org/articles/aix-00062/https://aixsci.org/articles/aix-00062/Researchers trained one neural network potential, called PFP, on their own database of quantum chemistry calculations, then used it instead of those calculations to study lithium movement, porous frameworks, alloy ordering and cobalt catalysts.Thu, 08 Oct 2026 21:45:39 GMThttps://aixsci.org/articles/aix-00061/https://aixsci.org/articles/aix-00061/Astronomers fine-tuned the open-source OpenCLIP model on labelled radio galaxy images and their descriptions, then used it to index about 170,000 extended radio sources so that a typed description or a sample picture returns similar objects.Thu, 08 Oct 2026 21:45:32 GMThttps://aixsci.org/articles/aix-00060/https://aixsci.org/articles/aix-00060/Researchers built a framework for training and sharing neural-network stand-ins for a widely used cosmology code, then used them to fit simulated sky data. The networks produced the predicted sky patterns in place of the original calculation.Thu, 08 Oct 2026 21:45:17 GMThttps://aixsci.org/articles/aix-00059/https://aixsci.org/articles/aix-00059/Researchers worked out what lies inside exoplanets from their mass and radius. A trained statistical stand-in replaced the slow physics model inside the sampler, and the same calculation ran in minutes instead of hours.Thu, 08 Oct 2026 21:45:10 GMThttps://aixsci.org/articles/aix-00058/https://aixsci.org/articles/aix-00058/Researchers trained a neural network to predict three combustion properties of fuels and their mixtures, then searched the model's internal representation for blends matching chosen targets. Eighty-six candidate mixtures came out; one blend of 22 components was put forward.Thu, 08 Oct 2026 21:45:04 GMThttps://aixsci.org/articles/aix-00057/https://aixsci.org/articles/aix-00057/Researchers built DARWEN, a genetic algorithm that searched for smaller versions of a large chemical reaction network used to model two hot Jupiter atmospheres, scoring each candidate against the full network.Thu, 08 Oct 2026 21:44:59 GMThttps://aixsci.org/articles/aix-00056/https://aixsci.org/articles/aix-00056/Researchers built DDMut-PPI, a neural network that estimates how much a change to one amino acid alters the strength with which two proteins stick together. The model was trained on measured binding energies and tested on mutations it had not seen.Thu, 08 Oct 2026 21:44:53 GMThttps://aixsci.org/articles/aix-00055/https://aixsci.org/articles/aix-00055/Astronomers trained small neural networks on 87 known polar ring galaxies, topped up with synthetic pictures, then used the models to score a catalogue of 315,000 galaxies. Three of the polar ring galaxies reported were found this way.Thu, 08 Oct 2026 21:44:46 GMThttps://aixsci.org/articles/aix-00054/https://aixsci.org/articles/aix-00054/Researchers built an open-source program, Careless, that turns raw X-ray reflection measurements into the quantities crystallographers need. A small neural network inside it works out how much each measurement must be rescaled, learning this from the recorded geometry of the experiment.Thu, 08 Oct 2026 21:44:41 GMThttps://aixsci.org/articles/aix-00052/https://aixsci.org/articles/aix-00052/Researchers ran AlphaFold-Multimer on 429 antibody-antigen complexes released after the model's training cut-off, then scored its predicted shapes against the experimentally determined ones. The AI did the predicting; the scoring was conventional software.Thu, 08 Oct 2026 21:44:15 GMThttps://aixsci.org/articles/aix-00051/https://aixsci.org/articles/aix-00051/Researchers treated bentonite clay with two kinds of ultrasound and used it to strip colour from soybean and sunflower oil. Seven machine learning models were fitted to the colour measurements, and the best one guided a search for better bleaching conditions.Thu, 08 Oct 2026 21:44:04 GMThttps://aixsci.org/articles/aix-00050/https://aixsci.org/articles/aix-00050/Researchers built a web server that scans a protein sequence for short helical stretches and predicts which ones bind nucleic acids. Eight machine learning classifiers, trained on physicochemical features, vote on each segment to produce the prediction.Thu, 08 Oct 2026 21:43:58 GMThttps://aixsci.org/articles/aix-00049/https://aixsci.org/articles/aix-00049/Researchers simulated gravitational waves from collapsing stellar cores under four descriptions of dense nuclear matter, then trained eight kinds of machine learning model to tell, from a waveform alone, which description produced it.Thu, 08 Oct 2026 21:43:52 GMThttps://aixsci.org/articles/aix-00048/https://aixsci.org/articles/aix-00048/Astronomers built CASTOR, an open-source tool that turns multi-band brightness measurements of an exploding star into a time series of synthetic spectra and a set of physical parameters. Gaussian Process regression does the filling-in.Thu, 08 Oct 2026 21:43:44 GMThttps://aixsci.org/articles/aix-00047/https://aixsci.org/articles/aix-00047/Researchers trained six small neural networks to reproduce an effective field theory model of how galaxies cluster, then used the trained stand-in, called EFTEMU, in place of the slower model code when fitting mock survey data.Thu, 08 Oct 2026 21:43:18 GMThttps://aixsci.org/articles/aix-00046/https://aixsci.org/articles/aix-00046/Researchers combined a fine-tuned machine-learning model of atomic forces with a graph neural network trained on quantum-chemistry data to estimate how the light-absorbing properties of mixed silver antiperovskites change with temperature.Thu, 08 Oct 2026 21:42:40 GMThttps://aixsci.org/articles/aix-00045/https://aixsci.org/articles/aix-00045/Researchers trained a latent stochastic differential equation model on simulated ten-year quasar observations. The network filled in gappy, noisy light curves and estimated the physical parameters behind them, replacing a Gaussian process fit used for comparison.Thu, 08 Oct 2026 21:42:33 GMThttps://aixsci.org/articles/aix-00044/https://aixsci.org/articles/aix-00044/Researchers trained a machine-learning model to predict how a gold electrode's electrons rearrange in response to nearby salt water, then used those predictions to supply the electric forces in a molecular dynamics simulation of the interface.Thu, 08 Oct 2026 21:42:25 GMThttps://aixsci.org/articles/aix-00043/https://aixsci.org/articles/aix-00043/Astronomers trained a three-dimensional neural network to sort a simulated universe into voids, walls, filaments and halos, then retrained it step by step to work on much sparser maps of matter.Thu, 08 Oct 2026 21:42:19 GMThttps://aixsci.org/articles/aix-00042/https://aixsci.org/articles/aix-00042/Researchers fitted principal component analysis and a convolutional autoencoder with an information-ordered bottleneck to simulated and real fast radio burst spectra. The learned representations produced the study's reconstructions, latent-space groupings and nine flagged outlier bursts.Thu, 08 Oct 2026 21:42:11 GMThttps://aixsci.org/articles/aix-00041/https://aixsci.org/articles/aix-00041/A team combed JWST and Hubble images of 11 nearby star-forming galaxies by eye, finding 292 candidate star clusters still wrapped in their birth dust. That human catalogue was then used to train image-recognition networks to pick out similar objects.Thu, 08 Oct 2026 21:41:48 GMThttps://aixsci.org/articles/aix-00040/https://aixsci.org/articles/aix-00040/Researchers fitted a probability model to records of 5,441 perovskite solar cells drawn from a public database. The same fitted model grouped the devices, predicted their efficiency, invented plausible new recipes and worked backwards to synthesis conditions for a chosen efficiency.Thu, 08 Oct 2026 21:41:41 GMThttps://aixsci.org/articles/aix-00039/https://aixsci.org/articles/aix-00039/Researchers trained a random forest classifier to tell exploding from non-exploding stellar cores, using labels from 100 two-dimensional supernova simulations, and let a neural network invent its own description of each star's interior.Thu, 08 Oct 2026 21:41:34 GMThttps://aixsci.org/articles/aix-00037/https://aixsci.org/articles/aix-00037/Archival Hubble and Webb images of the nearby dwarf galaxy WLM turned up a star that dims and brightens every 1.0934 days. A neural network trained to imitate a slow binary-star model let the team fit its light curve.Thu, 08 Oct 2026 21:35:21 GMThttps://aixsci.org/articles/aix-00036/https://aixsci.org/articles/aix-00036/Researchers simulated over four million powder X-ray diffraction patterns from 119,569 crystal structures, then trained 21 neural networks from scratch to read each pattern's crystal system and space group.Thu, 08 Oct 2026 21:35:12 GMThttps://aixsci.org/articles/aix-00035/https://aixsci.org/articles/aix-00035/Researchers prompted GPT-4 to pull measured band gap values out of sentences from the chemistry literature, then trained neural networks on the resulting dataset and compared them with models trained on existing collections.Thu, 08 Oct 2026 21:34:56 GMThttps://aixsci.org/articles/aix-00034/https://aixsci.org/articles/aix-00034/Researchers measured how stretched the orbits of 2465 Kepler planets and candidates are, then compared planets around stars belonging to two populations of the Milky Way's disk. A clustering model assigned each host star to one population or the other.Thu, 08 Oct 2026 21:34:47 GMThttps://aixsci.org/articles/aix-00033/https://aixsci.org/articles/aix-00033/Researchers rewrote three equations describing how lead-free solder deforms as neural networks whose internal weights are the equations' own coefficients. Training on published measurements, then refining with Bayesian statistics, produced values for those coefficients.Thu, 08 Oct 2026 21:34:38 GMThttps://aixsci.org/articles/aix-00032/https://aixsci.org/articles/aix-00032/Astronomers trained a convolutional neural network to convert noisy measurements of distorted galaxy shapes into maps of projected mass, using mock data matched to a coming wide-field survey, then applied it to real observations of the Coma cluster.Thu, 08 Oct 2026 21:34:22 GMThttps://aixsci.org/articles/aix-00023/https://aixsci.org/articles/aix-00023/Researchers trained a graph neural network, ProteinMPNN, to choose amino acid sequences for a given protein shape. The network wrote every sequence that was then made in bacteria and tested in the laboratory.Thu, 08 Oct 2026 09:24:28 GMThttps://aixsci.org/articles/aix-00031/https://aixsci.org/articles/aix-00031/Astronomers took near-infrared spectra of strong gravitational lenses and measured the distances of their background galaxies. The systems came from a candidate list assembled earlier by a neural network sifting through survey images.Wed, 07 Oct 2026 21:34:50 GMThttps://aixsci.org/articles/aix-00030/https://aixsci.org/articles/aix-00030/A machine learning filter picked out an exploding star in a galaxy 18.45 Mpc away and automatically requested a spectrum, which a robotic instrument began taking about seven minutes later.Wed, 07 Oct 2026 21:34:42 GMThttps://aixsci.org/articles/aix-00029/https://aixsci.org/articles/aix-00029/Researchers determined six structures of actin filaments in different chemical states at resolutions of 2.15–2.24 Å. A neural network located the filaments in the microscope images; everything downstream was conventional processing and hand-built modelling.Wed, 07 Oct 2026 21:34:36 GMThttps://aixsci.org/articles/aix-00028/https://aixsci.org/articles/aix-00028/Researchers fitted a neural network to quantum-mechanical calculations of gallium, then used it to run long atomic simulations. From these they computed the metal's phase diagram, its liquid structure, and how its crystals first form.Wed, 07 Oct 2026 21:34:26 GMThttps://aixsci.org/articles/aix-00027/https://aixsci.org/articles/aix-00027/Researchers trained ensembles of neural networks on simulated infrared spectra so that a measured spectrum of carbon monoxide on platinum could be turned into distributions of binding sites and local coordination.Wed, 07 Oct 2026 21:34:03 GMThttps://aixsci.org/articles/aix-00026/https://aixsci.org/articles/aix-00026/Researchers built a decision loop, ARROWS, that chooses which starting chemicals and temperatures to try when making a solid material. A neural network read the X-ray patterns after each heating step and told the loop which compounds had formed.Wed, 07 Oct 2026 21:33:57 GMThttps://aixsci.org/articles/aix-00025/https://aixsci.org/articles/aix-00025/Researchers trained five classifiers on 1200 experimentally made high-entropy alloys to predict which crystal structure a mixture would take. A random forest model predicted one new composition would be face-centred cubic; the alloy was then cast and X-rayed.Wed, 07 Oct 2026 21:33:45 GMThttps://aixsci.org/articles/aix-00024/https://aixsci.org/articles/aix-00024/Researchers stitched three bacterial enzymes into a library of 4,374 hybrids. Gaussian process models chose which hybrids to build and test across ten rounds, ending with a variant that made 54 ± 11 mg/L of fatty alcohols in E. coli.Wed, 07 Oct 2026 21:33:38 GMThttps://aixsci.org/articles/aix-00022/https://aixsci.org/articles/aix-00022/RELION builds three-dimensional maps of molecules from noisy electron microscope images. A statistical model learns the map, the signal and the noise from the data themselves, replacing settings a user would otherwise choose by hand.Wed, 07 Oct 2026 21:33:01 GMThttps://aixsci.org/articles/aix-00021/https://aixsci.org/articles/aix-00021/Astronomers trained a convolutional neural network on low-resolution starlight from the LAMOST survey, using sharper measurements from another survey as answers, then used it to produce a catalogue of parameters and abundances for 1,210,145 giant stars.Wed, 07 Oct 2026 21:32:29 GMThttps://aixsci.org/articles/aix-00020/https://aixsci.org/articles/aix-00020/Astronomers trained a deep learning classifier, ExoMiner, to tell real planets from look-alike signals in Kepler telescope data. Scores from the network, filtered by a threshold and catalogue checks, produced 301 newly validated exoplanets.Wed, 07 Oct 2026 21:32:18 GMThttps://aixsci.org/articles/aix-00019/https://aixsci.org/articles/aix-00019/Researchers catalogued the transient noise recorded by the two Advanced LIGO detectors during their first three observing runs. A convolutional neural network, trained on images labelled by experts and volunteers, sorted each noise blip into a class and gave a confidence score.Wed, 07 Oct 2026 21:32:11 GMThttps://aixsci.org/articles/aix-00018/https://aixsci.org/articles/aix-00018/Astronomers put a random-forest classifier on the live stream of transient alerts from a sky survey. Each night it flagged the ten objects it was least sure about, those were observed with a telescope, and the answers went back into its training.Wed, 07 Oct 2026 21:32:04 GMThttps://aixsci.org/articles/aix-00017/https://aixsci.org/articles/aix-00017/Astronomers built BTSbot, a neural network that scores each alert from the Zwicky Transient Facility for whether it marks a bright transient. In production it saved 296 sources and requested spectra automatically, work previously done by human scanners.Wed, 07 Oct 2026 21:31:49 GMThttps://aixsci.org/articles/aix-00016/https://aixsci.org/articles/aix-00016/Researchers searched a space of 797,504 possible nickel-titanium alloy recipes for ones with low thermal hysteresis. Models trained on measured alloys predicted each candidate's value and its uncertainty, and those predictions chose all 36 recipes that were actually made.Wed, 07 Oct 2026 21:31:42 GMThttps://aixsci.org/articles/aix-00015/https://aixsci.org/articles/aix-00015/Researchers built Uni-MOF, a transformer trained first on hundreds of thousands of porous crystal structures and then on around 3,000,000 adsorption measurements, so that a structure file plus a gas, temperature and pressure yields a predicted uptake.Wed, 07 Oct 2026 21:31:32 GMThttps://aixsci.org/articles/aix-00013/https://aixsci.org/articles/aix-00013/Researchers X-rayed lithium-ion cathodes and used an image-recognition network to outline every active particle in the three-dimensional scans, then measured how far each had pulled away from the conductive matrix around it.Wed, 07 Oct 2026 21:29:07 GMThttps://aixsci.org/articles/aix-00012/https://aixsci.org/articles/aix-00012/Researchers fitted a machine-learning model of the forces between silicon and oxygen atoms, trained on quantum-mechanical calculations chosen by the model's own uncertainty, then used it to simulate silica under pressure, porous structures and amorphous silicon monoxide.Wed, 07 Oct 2026 21:28:44 GMThttps://aixsci.org/articles/aix-00011/https://aixsci.org/articles/aix-00011/Researchers let a Bayesian active-learning system called CAMEO choose which germanium–antimony–tellurium compositions to measure by X-ray diffraction. Over 19 cycles it mapped the material's phases and picked out a composition averaging Ge4Sb6Te7.Wed, 07 Oct 2026 21:27:53 GMThttps://aixsci.org/articles/aix-00010/https://aixsci.org/articles/aix-00010/Researchers built a laboratory in which machine-learning models drew up recipes for making new solid materials from published literature, robots carried them out, and further models read the X-ray measurements that showed what had formed.Wed, 07 Oct 2026 21:26:58 GMThttps://aixsci.org/articles/aix-00009/https://aixsci.org/articles/aix-00009/Researchers measured how 218 stitched-together channelrhodopsin proteins behaved in human cells, then trained Gaussian process models on those measurements to predict, out of 118,098 possible variants, which untested ones would reach the cell membrane and which to build next.Wed, 07 Oct 2026 21:26:39 GMThttps://aixsci.org/articles/aix-00008/https://aixsci.org/articles/aix-00008/Researchers built crYOLO, a convolutional neural network that scans cryo-electron microscopy images and marks the positions of individual protein particles, replacing a step normally done by hand or by conventional picking algorithms.Wed, 07 Oct 2026 21:26:21 GMThttps://aixsci.org/articles/aix-00007/https://aixsci.org/articles/aix-00007/AlphaFold, a neural network trained on known protein structures, predicts where every heavy atom in a protein sits using only its amino acid sequence and alignments of related sequences. The predicted structures are the result itself.Wed, 07 Oct 2026 21:25:19 GMThttps://aixsci.org/articles/aix-00006/https://aixsci.org/articles/aix-00006/Researchers fitted a statistical model to the amino acid patterns in families of related protein sequences, letting it work out which residues are coupled. Those couplings became distance constraints that folded 15 test proteins into three-dimensional shapes.Wed, 07 Oct 2026 21:25:13 GMThttps://aixsci.org/articles/aix-00005/https://aixsci.org/articles/aix-00005/Researchers imaged relaxed mouse cardiac muscle by cryo-electron tomography and built an atomic model of its thick filament. Learned software traced the filaments, cleaned up images, and predicted the shapes of the proteins fitted into the density.Wed, 07 Oct 2026 21:25:06 GMThttps://aixsci.org/articles/aix-00004/https://aixsci.org/articles/aix-00004/Researchers built a self-driving laboratory in which a statistical learning model chose which enzyme sequences to build, and robots assembled and tested them. Over twenty rounds, four independent agents each found designs more heat-stable than the natural starting proteins.Wed, 07 Oct 2026 21:24:22 GMThttps://aixsci.org/articles/aix-00003/https://aixsci.org/articles/aix-00003/A team assembled a complete structural model of Commander, a protein machine in cells linked to Ritscher-Schinzel syndrome, by combining X-ray crystallography and cryo-electron microscopy with AlphaFold2 Multimer predictions that filled in the parts experiments did not resolve.Wed, 07 Oct 2026 21:24:12 GMThttps://aixsci.org/articles/aix-00002/https://aixsci.org/articles/aix-00002/Researchers built ARTINA, a workflow that takes NMR spectra and a protein's sequence and returns a structure. Five trained models read the spectra, fill in missing measurements and choose between candidate structures, doing work that normally needs an expert's eye.Wed, 07 Oct 2026 21:23:35 GMThttps://aixsci.org/articles/aix-00001/https://aixsci.org/articles/aix-00001/A new laboratory assay produced around 776,000 measurements of how firmly protein domains hold their shape. Learned and fitted models chose the domains, generated some of the test sequences, turned sequencing counts into stability numbers and helped interpret the results.Wed, 07 Oct 2026 15:08:56 GMT