ML Scientist, Molecular Design Platform

Abiologics; Watertown (MA); On-site; USD 132,000 - 259,000; Full time

Benefits offered by this job

Healthcare coverage

Annual incentive program

Retirement benefits

Job summary

Abiologics, Inc. is a pioneering biotechnology company hiring a Machine Learning Scientist to train, validate, and ship predictive models powering its molecular design platform.

You will build models scoring molecules for developability (ADMET), potency, and synthesizability, while collaborating with chemists, biologists, and engineers to bring models into production. In the first year, expect two to three models in production with documented benchmarks and clear applicability domains.

Qualifications

• PhD in computational chemistry or related field, or MS + 3–6 years of hands-on modeling.
• Experience training and validating ML models on chemical data.
• Experience with ADMET/developability, potency, or synthesizability is preferred.

Responsibilities

• Train and maintain ML models for ADMET and developability endpoints.
• Build potency and binding models to rank molecules.
• Develop synthesizability models and retrosynthesis-informed filters.
• Apply structure-aware models (GNNs, transformers on molecules).
• Curate assay data, define benchmarks, quantify uncertainty and applicability domain.
• Package models for the platform with versioning and drift monitoring.
• Present results and limits of each model to chemists and project teams.

Skills

Python

PyTorch

ML models

Communication

Tools

RDKit

GNNs

Transformers

Job description

Abiologics, Inc. is a pioneering biotechnology company hiring a Machine Learning Scientist to train, validate, and ship predictive models powering its molecular design platform.

You will build models scoring molecules for developability (ADMET), potency, and synthesizability, while collaborating with chemists, biologists, and engineers to bring models into production. In the first year, expect two to three models in production with documented benchmarks and clear applicability domains.