(Senior) Machine Learning Scientist

Abiologics•Watertown, MA

6h•$132,000 - $258,500

About The Position

Abiologics, Inc. is hiring a Machine Learning Scientist to train, validate, and deploy predictive models for its molecular design platform. The models will score chemical molecules for developability (ADMET), potency, synthesizability, and 3D structure. The scientist will work alongside chemists, biologists, and platform engineers. Success in the first year means having two or three models in production use by project teams, with documented benchmarks, calibrated uncertainty, and a clearly defined applicability domain.

Requirements

PhD in computational chemistry, cheminformatics, machine learning, or a related field, or an MS plus equivalent industry experience (about 3–6 years of hands-on modeling).
Demonstrated experience training and validating machine learning models on chemical or molecular data in industry or academia, with shipped or published results.
Direct experience with at least two of the following: ADMET/developability, potency/binding, synthesizability/retrosynthesis, or protein or ligand structure.
Fluency with molecular representations such as fingerprints, descriptors, graphs, and 3D, and tools such as RDKit.
Strong Python skills and experience with a deep learning framework; PyTorch is preferred. Experience with GNNs or transformers on molecules.
Rigorous evaluation practices, including scaffold or temporal splits, baselines, noisy-label handling, and uncertainty estimation.
Clear communication with chemists and biologists.

Nice To Haves

Publications, open-source contributions, or competition results in molecular machine learning.
Experience with generative design, active learning, or Bayesian optimization loops.
Experience with large-scale training, MLOps, and model serving.
Familiarity with docking, free-energy methods, or protein structure prediction tools.
Experience working with biologics or non-small-molecule modalities.

Responsibilities

Train and maintain machine learning models for ADMET and developability endpoints, including solubility, permeability, metabolic stability, toxicity, and aggregation risk.
Build potency and binding models, including QSAR and ligand- or structure-based models, to rank and prioritize molecules against targets.
Develop synthesizability models, including synthetic accessibility scoring, reaction-outcome prediction, and retrosynthesis-informed filters.
Apply and fine-tune structure-aware models, including GNNs, equivariant networks, protein-ligand structure, and docking-informed machine learning.
Curate and clean assay data; define scaffold- or time-split benchmarks; and quantify uncertainty and applicability domain.
Package models for the platform in collaboration with engineers, ensuring they are versioned, reproducible, and monitored for drift.
Present model results and limitations to chemists and project teams, and incorporate their feedback into retraining.