Marwan El Saabi - Data Analyst | Python, SQL & Machine Learning

Retail demand forecasting. 4.8% better than the strongest baseline.

Weekly SKU-level forecasts on more than a million real e-commerce transactions. A global LightGBM model, benchmarked against naive, seasonal-naive and moving-average baselines under rolling-origin backtesting, with every feature read at the forecast origin to rule out leakage.

Try the live app View the code

Retail Demand Forecasting app: backtest metrics and model comparison table — Backtest results, averaged over every fold and SKU.

AI drug designer. An LLM proposes. RDKit validates.

Gemini generates candidate molecules as SMILES strings, a cheminformatics validation engine filters out physically impossible structures, and the survivors are rendered as optimized 3D models. Built with Python, Streamlit, the Gemini API and RDKit.

Try the live app View the code

AI drug designer results: five generated molecules validated by RDKit, the first shown as a 2D structure and a 3D model

More work. Genomics, computer vision and drug screening.

RNA-seq analysis dashboard.

Differential expression with Welch's t-test and Benjamini-Hochberg correction, PCA, volcano and MA plots and heatmaps, on synthetic data and the real airway dataset (Himes et al., 2014).

Live app Code

Try it here RNA-seq dashboard with volcano and MA plots

Alzheimer's drug classifier.

Virtual screening on curated ChEMBL IC50 data with Morgan fingerprints. Random Forest led four benchmarked models with 0.96 AUROC under stratified 5-fold cross-validation.

Code

ROC curves comparing KNN, SVM, Random Forest and XGBoost

Melanoma detection, without deep learning.

Otsu, K-means and watershed segmentation selected by IoU, then 40 shape, colour and texture features and an SVM. 90.2% accuracy and 84% melanoma sensitivity on ISIC 2018.

Live app Code

Segmentation example: original lesion, ground truth mask and automatic segmentation

I trained as a biologist, but the part I always enjoyed most was the data. Biostatistics was my favourite subject, and my undergraduate thesis turned photographs of salamanders into a number that could be measured and compared, work that later became part of a peer-reviewed paper. That's what took me to the MSc in Bioinformatics, and to building projects well beyond biology.

The toolkit.

Languages

Python: NumPy, Pandas, Scikit-learn, OpenCV, Matplotlib
SQL: Relational modelling, ER design, querying
R / Bioconductor: Tidyverse, DESeq2, ggplot2
MATLAB: Neural networks, SVMs and image processing

Modelling

Machine learning: LightGBM, XGBoost, model benchmarking, cross-validation
Time series forecasting: Rolling-origin backtesting, demand models, wMAPE / MASE
Generative AI: LLM APIs (Google Gemini), generation with rule-based validation

Tools and infrastructure

Streamlit: Interactive data applications
Git / GitHub: Version control, GitHub Actions
Nextflow: Bioinformatics pipeline orchestration
Linux and HPC: Bash, Docker, Conda, SLURM on the FinisTerrae III supercomputer (CESGA)

Education and research.

2025-2027
In progress

MSc in Bioinformatics Applied to Health Sciences

University of A Coruña

2026
Publication

Melanism in salamanders: effects of insularity, sex and size on dorsal colouration in Salamandra salamandra

Alarcón-Ríos, L., Saabi González, M. and Velo-Antón, G. Journal of Zoology

2024

BSc in Biology (Mention: Biotechnology)

University of Vigo

2026
Training badge

AWS Cloud Quest: Cloud Practitioner

Amazon Web Services Training and Certification

2025
Certification

English B2 (CertAcles)