PHILIPPE MEYER

Codes

Here are the codes I have developed or co-developed. They range from Python packages to repositories supporting my papers, mainly in cheminformatics and synthetic biology. Each name links to its GitHub repository, and the associated paper, when there is one, is linked after the description.

  • 2026 – Agentic Retrobiosynthesis: MorganBioPilot, a retrobiosynthesis framework in which a large language model agent (Qwen2.5-7B-Instruct, prompted or LoRA fine-tuned) chooses which molecule to expand next, compared with breadth-first, depth-first, greedy best-first and Monte Carlo tree search (UCT) on a deterministic engine in ECFP fingerprint space. Paper Fields: Retrobiosynthesis · Synthesis planning · LLM agents · LLM fine-tuning · Tree search Tools: RDKit · PyTorch · Hugging Face TRL · PEFT · Qwen2.5 · vLLM · Slurm (GPU cluster)
  • 2026 – Vectorized Chemical Reactions: morganrxn, a Python package representing chemical and enzymatic reactions as vectors in ECFP fingerprint space, both for fast applicability filtering of reaction rules and as features for machine learning classifiers (logistic regression, random forest, gradient boosting, extra-trees, MLP) predicting reaction classes and EC numbers. Paper Fields: Reaction informatics · Reaction rules · Atom mapping · Enzyme function prediction · Reaction classification Tools: RDKit · RXNMapper · scikit-learn · SciPy · NumPy · pandas · Slurm (HPC cluster)
  • 2026 – Metabolic Biosensors: COVID-19 prognosis and diagnosis from the full growth curves of engineered E. coli biosensor strains, with machine learning (SVM, logistic regression, XGBoost), deep learning time-series models (1D CNN, temporal convolutional networks) with FiLM conditioning and multi-strain fusion, and generalized additive mixed models (GAMMs). Paper Fields: Microbial biosensors · Clinical diagnostics · Time-series classification · Hyperparameter optimization · Mixed models Tools: PyTorch · scikit-learn · XGBoost · Optuna · Ray · mgcv (R) · Matplotlib
  • 2026 – Bacterial Reservoir Computing: Living E. coli used as a reservoir computer: a neural network encodes input features as growth media, ridge regression reads out the bacterial growth, and an artificial metabolic network (AMN), a hybrid neural–mechanistic model, predicts growth in silico, with SVM, MLP and XGBoost as baselines and for COVID-19 patient classification. Paper Fields: Synthetic biology · Microbial biocomputing · Genome-scale metabolic models · Flux balance analysis · Hybrid mechanistic–ML models Tools: TensorFlow / Keras · COBRApy · scikit-learn · XGBoost · SciPy · pandas
  • 2025 – Deterministic Molecular Reverse Engineering: molecule-signature, a Python package available on conda-forge that recovers molecular structures from their ECFP (Morgan) fingerprints by learning-free deterministic enumeration based on molecular signatures and linear Diophantine systems. Paper, Application Fields: Cheminformatics · Molecular fingerprints · Inverse QSAR · Molecular enumeration · Combinatorics Tools: RDKit · SymPy · NetworkX · SciPy · NumPy · pandas
  • 2025 – Generative Molecular Reverse Engineering: A generative encoder–decoder transformer that produces molecular structures (SMILES) from ECFP fingerprints, pre-trained on eMolecules and fine-tuned on MetaNetX metabolites, decoded with beam search and benchmarked against deterministic enumeration. Paper Fields: Generative chemistry · De novo molecular design · Chemical language models · Transfer learning Tools: PyTorch · Lightning · TorchMetrics · RDKit · SentencePiece · TensorBoard · pandarallel
  • 2025 – Linear Diophantine Systems: lineardiophantine, a Python package that solves linear Diophantine systems over the natural numbers with the Contejean–Devie algorithm, and over the integers with the Smith normal form. Fields: Computational algebra · Integer linear systems · Combinatorics Tools: NumPy · hsnf · pytest · pytest-cov
  • 2024 – Glyph Similarities: A siamese convolutional neural network, trained on Omniglot with data augmentation, that measures similarities between glyphs and writing systems, followed by multidimensional scaling and Ward hierarchical clustering of 51 historical European, Mediterranean and Middle Eastern scripts. Paper, 3D view Fields: Computer vision · Metric learning · Few-shot learning · Clustering · Digital humanities Tools: TensorFlow / Keras · OpenCV · scikit-image · scikit-learn · SciPy · Plotly · Pillow
  • 2023 – Human Body Shapes: Topological data analysis of 3D human body scans, with persistence diagrams compared through bottleneck and Wasserstein distances and vectorized as persistence silhouettes, for anomaly detection and body shape classification. Paper Fields: Topological data analysis · 3D shape analysis · Anomaly detection · Clustering · Anthropometry Tools: GUDHI · POT · scikit-learn · SciPy · meshio · Matplotlib · pytest
  • 2022 – Anthropometric Data Explorer: ExploRed, an R Shiny application for exploring the French national sizing survey (Campagne Nationale de Mensuration, 2004–2006) and its ~11 600 3D body scans, served from sodium-encrypted data unlocked by a password acting as the private key. It provides descriptive statistics, correlation and hypothesis testing, interactive trivariate scatter plots, body scans rendered as point clouds or meshes, and survey calibration of measurement means on INSEE census margins over 2001–2021, all stratified by sex, age class, region and socio-professional category, and is deployed with Docker on an internal server for IFTH and UTT users. Fields: Survey statistics · Survey calibration · Anthropometry · Data visualization · Data security Tools: R Shiny · icarus · plotly · ggplot2 · rstatix · sodium · Docker
    Watch a demo of the application