materials-chemistry/ai in a supporting role/Scientific Reports 2023 · v2
Neural network predicts yield of chitosan nanoparticles grown with olive leaf extract
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.
spectrum · one line per step, placed by what the step does · bright lines used AI
Artificial intelligence-based optimization for chitosan nanoparticles biosynthesis, characterization and in‑vitro assessment of its anti-biofilm potentiality
Scientific Reports, 2023
doi:10.1038/s41598-023-30911-6 · record aix-00103 v2 · checked 2026-10-08
- AI was for
- Property prediction
- Model family
- Multilayer perceptron
- Checked by
- Experimental
- Code
- not reported
AI processed or interpreted data, but the main finding does not rest on it.
What this research was about
Chitosan is a sugar-based material obtained from the shells of crabs, prawns and other crustaceans. When it is broken down into particles only a few billionths of a metre across, it behaves differently from the bulk material, and such particles are of interest for acting on bacteria and fungi. One way of making them is to mix a chitosan solution with a plant extract, letting the chemicals in the extract drive the particles into shape. The difficulty is that the result depends on several conditions at once: how concentrated the chitosan is, how much extract is added, how acidic the mixture is, how warm it is, and how long it is left. These settings interact, so changing one alters the best value of another, and testing every combination is impractical.
The researchers prepared chitosan nanoparticles with an extract of leaves from Olea europaea, the olive tree, and used a structured plan of 50 experiments that varied those five conditions together. They measured the amount of nanoparticle material produced in each run, looked for the combination that gave the most, and then examined the resulting particles with microscopes and spectroscopic instruments. They also tested the particles against biofilms, the slimy mats that bacteria and fungi build on surfaces, using Pseudomonas aeruginosa, Staphylococcus aureus and Candida albicans.
Where AI came in
The artificial intelligence here was an artificial neural network, a flexible mathematical model that learns a relationship between inputs and an output by adjusting internal weights until its predictions match the measurements it is shown. The researchers built one in the commercial software JMP Pro 14, feeding it the five process conditions through five input units, passing these through a single hidden layer of 20 units, and reading out one number: the predicted nanoparticle yield in milligrams per millilitre. The 50 measured runs were split into portions for training, for checking progress and for testing, and settings such as the number of units and the learning rate were chosen by trial and error.
The network stood in for a conventional statistical formula. Alongside it the team fitted a quadratic response surface model, the standard curve-fitting approach for experiments of this kind, and compared the two by how closely each matched the measured yields. Choosing the final recipe was done with a desirability function applied to the fitted models rather than by the network itself, and the chosen conditions were then confirmed in the laboratory: the network's predicted yield was 20.21 mg/mL and the measured yield 21.15 mg/mL. The particle characterisation and the biofilm tests did not involve the network.
Written by AIxSci from the checked record below, to give context for readers outside the field. It is not part of the record.
The work
Technical · from the record
Chitosan nanoparticles were biosynthesised by mixing chitosan solution with Olea europaea leaves extract, and a face-centred central composite design of 50 experiments varied chitosan concentration, extract concentration, initial pH, temperature and incubation time. A neural network with five input neurons, a hidden layer of 20 neurons and one output neuron was fitted to those 50 runs to predict yield, and its R, RASE and AAE were compared with the quadratic response surface model. At the conditions selected by the desirability function (1% chitosan, 100% extract, pH 4.47, 53.83 °C, 60 min) the network predicted 20.21 mg/mL and the measured yield was 21.15 mg/mL. The particles were 6.91 to 11.14 nm by TEM with a zeta potential of 33.1 mV, and reduced biofilm formation, metabolic activity, protein and exopolysaccharide content and hydrophobicity of P. aeruginosa, S. aureus and C. albicans over 10 to 1500 µg/mL.
How AI was used
The 50-run face-centred central composite design matrix and the measured nanoparticle yields were used as the dataset for an artificial neural network built in JMP Pro 14. The data were divided into training, validation and testing portions, with the input layer taking the five process variables as five neurons, a hidden layer of 20 neurons, and a single output neuron for nanoparticle yield in mg/mL. Network settings were chosen by trial and error over the number of neurons, holdback ratio and learning rate, with the reported configuration using 5000 tours, an NTanH model with 20 nodes, a learning rate of 0.1 and a holdback validation fraction of 0.2; training continued until RMSE, MAD and SSE were lowest and R highest for training and validation. The fitted network was then run over the design points to produce predicted yields and residuals, and its R, root average squared error and average absolute error were compared with those of the second-order polynomial response surface model fitted to the same data. The desirability function was applied to obtain optimum conditions, for which a network-based predicted yield was reported and a confirmation synthesis was then performed.
The shape of the work
Structural · the record, drawn
no AI
Run 50-trial face-centred central composite design of CNP biosynthesis
Physical execution, by hand or by robot.
The experimental design used in this study consisted of 50 experimental trials including 8 trials at the central point.where the paper describes this · verbatim
no AI
Fit second-order response surface model to the design data
Fitting model parameters, including fine-tuning an existing model.
The theoretical relationships among the independent variables and the outcomes (CNPs biosynthesis, mg/mL) were identified by applying the polynomial equationwhere the paper describes this · verbatim
AI
Train artificial neural network on the design matrix
Fitting model parameters, including fine-tuning an existing model. The AI stood in for statistical model.
FCCCD matrix, and experimental data (Table 1), were subjected to ANN analysis.where the paper describes this · verbatim
AI
Predict CNP yield with the network and compare errors against the response surface model
Running a trained model over new data to predict, classify or score. The AI stood in for statistical model.
The predicted values of CNPs biosynthesis by ANN corresponding to each experimental result were given in Table 1.where the paper describes this · verbatim
no AI
Select optimum synthesis conditions with the desirability function
Iterative search over a space.
Using the desirability function, the optimum conditions for maximum CNP biosynthesis with Olea europaea was determined theoretically and verified experimentally.where the paper describes this · verbatim
no AI
Verify the predicted optimum by biosynthesis experiment
Testing outputs against ground truth.
Under the previous conditions, the maximum experimental value of CNPs yield using Olea europaea leaves extract was 21.15 mg/mL.where the paper describes this · verbatim
no AI
Characterise the nanoparticles and assay antibiofilm activity
Physical execution, by hand or by robot.
By employing microdilution method, the inhibitory effect of chitosan-NPs with different concentrations (10, 20, 50, 100, 200, 500, 1000, 1500 μg/mL) was assessed.where the paper describes this · verbatim
What the record says
Technical · every part carries its own basis
+ in the paper~ our reading− not reported
How to read the quotations. A quotation shows where the paper describes something. It does not quote every value beside it: one passage locates a part of the work, and values without their own quotation are our reading of that passage.
The neural network modelled and predicted nanoparticle yield from the 50-run design, but the synthesis, the characterisation and the antibiofilm results do not depend on it; the optimum conditions were produced by a desirability function over the fitted models and then confirmed experimentally
The ANN architecture is composed of an input layer with the five independent factors (five neurons)where the paper describes this · verbatim
The theoretical predicted value of CNPs biosynthesis by ANN (20.21 mg/mL) was considerably closer to the experimental value (21.15 mg/mL)where the paper describes this · verbatim
What this paper did not report
Technical · absence is published deliberately
- CodeWhether the code is available is not stated.
- Trained model weightsWhether the trained model is available is not stated.
- DataWhether the data are available is not stated.
- ComputeThe hardware or time used is not stated.
- How many were testedThe paper gives no count of what was tested.
- Version of Artificial neural network (JMP Pro 14 neural platform, NTanH with 20 nodes)Which version of the model was used is not stated.
About this article
Record aix-00103, version 2, checked by a person on 2026-10-08. The record describes the paper; it does not assess whether the paper's findings are right. The paper is published under CC-BY; quotations are at most 25 words. How we work · Report an error