Machine Learning Applications | Dr. Mohamed Elsenety

Universal pH estimation via Deep Artificial Neural Networks directly from smartphone photos of 1-block, 2-block, and 4-block indicator strips under variable ambient illumination.

Inputs: Live Camera Capture or Upload → Auto color extraction
Model: Deep Neural Network (ANN) trained on RGB color spaces
Output: Quantitative pH value with confidence index

Deep ANN Android App Computer Vision Google Play RGB Stats

Project Page Google Play Store

High-accuracy prediction of solar cell Power Conversion Efficiency (PCE%) for imidazole-based organic sensitizers using molecular fingerprints and multilayer perceptron neural networks.

Inputs: Chemical SMILES string or molecular sketch
Algorithm: MLP Regressor (Deep Neural Net) + Daylight Fingerprints
Citation: Sci Rep 14, 25890 (2024) · DOI: 10.1038/s41598-024-76739-6

MLP Regressor Streamlit SMILES Daylight Fingerprint Nature Sci Rep

Launch App Model Guide Read Paper

High-throughput deep learning prediction of D-π-A organic dye absorption maxima (λmax) when anchored to titanium dioxide surfaces, accelerating photocatalyst and solar cell design.

Inputs: SMILES string notation of D-π-A chromophore
Model: Deep Learning Artificial Neural Network (ANN)
Citation: ChemistrySelect (2025) · DOI: 10.1002/slct.202501556

Deep ANN Titania Interface Streamlit Absorption Maxima ChemistrySelect

Launch App Read Paper

Ultra-fast solvatochromic prediction of D-π-A organic dye absorption peaks across diverse liquid media using an advanced ensemble deep learning framework.

Inputs: SMILES string & solvent dielectric parameters
Model: Ensemble Deep Learning Neural Architecture
Citation: Spectrochimica Acta A (2024) · DOI: 10.1016/j.saa.2024.125536

Ensemble Deep Learning Solvatochromism Streamlit Elsevier SAA

Launch App ScienceDirect Paper

Predict photovoltaic performance parameters (PCE%, Voc, Jsc, FF) of Triphenylamine (TPA) organic dyes from SMILES representations using multi-target deep neural networks with R² > 0.90.

Inputs: SMILES string or batch molecular upload
Accuracy: R² = 0.90 (Jsc), 0.91 (Voc), 0.90 (FF)
Status: Manuscript under peer review · Demo live

TPA Dyes Multi-Output ANN Streamlit Batch Screening

Launch Streamlit App Guide & Details

Facile and highly precise pH-value estimation using ubiquitous standard commercial pH test paper based on machine learning algorithms and everyday mobile devices.

Inputs: Mobile camera snapshot or gallery image
Algorithm: K-Nearest Neighbors (KNN) classifier / regressor
Citation: Sci Rep 12, 22617 (2022) · DOI: 10.1038/s41598-022-27054-5

KNN Machine Learning Mobile Sensor Nature Sci Rep Colorimetry

Launch App Read Paper

Computational Workflow

From Chemical Structure to Predictive Intelligence

How our AI models translate molecular SMILES and optical inputs into reliable physical properties in milliseconds.

Molecular Encoding

Chemical structures are parsed via RDKit into canonical SMILES and converted into high-dimensional Daylight, Morgan, or MACCS bit vectors.

Deep Neural Inference

Trained multilayer perceptrons (MLPs) and ensemble neural architectures evaluate complex non-linear electronic interactions without costly DFT runs.

Actionable Predictions

Researchers receive instantaneous predictions for PCE%, absorption maxima, or pH metrics to prioritize synthetic candidate targets in the wet lab.

Machine Learning Apps

Smart_pH Universal Estimator

Imidazole Dyes in DSSCs (PCE%)

Dye Max.Abs. on TiO2 Surfaces

Dye Max.Abs. in Diverse Solvents

TPA-Based Dyes in DSSCs

Fast pH Estimator (Colorimetric)

From Chemical Structure to Predictive Intelligence

Molecular Encoding

Deep Neural Inference

Actionable Predictions