Matt Ebrahim
Senior Data Scientist specializing in AI for Drug Discovery
LinkedIn
GitHub
Scholar
Resume
I'm a Senior Data Scientist at Formation Bio, where I lead drug repurposing initiatives using machine learning. I build scalable models—transformers, graph neural networks, diffusion models—for molecule generation, ADMET prediction, and clinical portfolio prioritization.
I also teach graduate-level AI courses at Northeastern University, focusing on deep learning applications in healthcare and drug discovery.
Lead drug repurposing with GNN pipelines on biomedical knowledge graphs. Build MRI surrogate endpoint models and fine-tune domain-specific LLMs for ontology mapping.
Led generative AI for de novo molecular design. Developed multimodal GNN architecture for BBB permeability (NeurIPS 2025 submission). Delivered molecules progressing through in vitro/in vivo validation.
Built ADMET prediction models and molecular property optimization pipelines. Contributed to multi-objective drug design workflows integrating ML with medicinal chemistry.
Deep learning for medical imaging. CycleGAN for MRI synthesis from CT. CNN-MLP for aortic flow estimation from wearable sensors (first-author publication + U.S. patent).
ML models for terahertz imaging and burn injury diagnostics. 8+ first-author and 30+ co-authored peer-reviewed publications.
Generative AI
Diffusion (DDPM), Transformers (GPT, T5), VAE, GAN, RNN/LSTM, Reinforcement Learning
Graph ML
Graphormer, GCN, GAT, MPNN, PyG, DGL, Knowledge Graphs
Drug Discovery
RDKit, DeepChem, AutoDock Vina, ESMFold, AlphaFold, ADMET modeling
Foundation Models
BioMistral, BioMegatron, SapBERT, MedGemma, BioMedCLIP, DINOv3
ML Frameworks
PyTorch, TensorFlow, Hugging Face, Scikit-learn, Optuna
Cloud & MLOps
Azure ML, AWS (Bedrock, SageMaker), GCP, Snowflake, Docker
Teaching — Northeastern University
BioMap: Biomedical Entity Linking
Benchmarking framework for biomedical entity linking using SapBERT, BioMegatron, and LLMs (GPT-4, Gemini) on MedMentions dataset.
DiffBind: Diffusion Models for Binding Affinity
DDPM implementation for predicting protein-ligand binding affinity using equivariant diffusion and the BindingMOAD dataset.
Catflix: AI-Generated Cat Videos
Automated pipeline that generates and uploads animated videos for cats to YouTube, featuring procedurally generated bug animations.
pdf2audiobook: PDF to Audiobook Converter
Convert any PDF into a chapter-aware audiobook with streaming pipeline. Includes web interface for real-time playback and read-along.
Multimodal Graph-Attention Networks with QM-Guided Cross-Attention for ADMET Prediction
Deep Learning for Aortic Flow Estimation from SCG
