Computational Drug Discovery AI Model Engineer

Position Title: Computational Drug Discovery AI Model Engineer

Length of Contract: 4 months (September-December)

Remote

What are the top 3-5 skills, experience or education required for this position:

1. Strong experience in machine learning, computational chemistry, cheminformatics, or related scientific data science domains

2. Experience in training and fine-tuning models on domain-specific scientific datasets

3. Experience with 3D modeling workflows, pose generation, or docking-based comparison studies

4. Working knowledge of benchmarking methodologies and model performance evaluation

5. Python and/or related scripting and automation experience

Project: Build and benchmark fine-tuned co-folder and affinity prediction workflows for LO projects

We are seeking a contractor to support the development, fine-tuning, and benchmarking of AI/ML models for computational drug discovery applications across multiple lead optimization programs. This role will help accelerate model development and evaluation by creating a scalable workflow that can start from protein and ligand sequence and produce a 3D structure and affinity prediction with minimal manual intervention.

The contractor will work closely with Computational Drug Discovery scientists to compile client project data, train and evaluate models, and establish reproducible benchmarking across multiple LO projects.

Responsibilities

• Start from a pre-trained foundation model such as Boltz-2 or AISB and fine-tune it on client data.
• Compile client datasets from multiple advanced LO projects including both experimental protein-ligand complex structures and potency/affinity data.
• Evaluate dataset variation across projects to challenge model applicability domains, ranging from small sets with a few structures and ~100 affinity data points to larger sets with 10+ structures and thousands of affinity data points.
• Benchmark model performance against established methods and tools, including DeepAutoQSAR, GatorAffinity, MPNN, MM-GBSA, FEP, and related approaches.
• For 3D workflows, compare pose generation using co-folder approaches against docking-based methods and assess performance.
• Use project-specific thresholds to evaluate recall and precision as primary metrics.
• Using confidence and domain of applicability, develop model ranking across affinity prediction model types
• Determine whether a single fine-tuned co-folder model trained across selected projects is sufficient, or whether project-specific models are required.
• Design the workflow with automation in mind, ideally enabling an end-to-end process from protein and ligand sequence to predicted 3D structure and affinity.
• Optimize workflow for efficiency that single predictions can be generated in less than one minute on GPU or CPU resources.

Required Skills and Experience

• Strong experience in machine learning, computational chemistry, cheminformatics, or related scientific data science domains.
• Experience in training and fine-tuning models on domain-specific scientific datasets.
• Working knowledge of benchmarking methodologies and model performance evaluation.
• Experience with 3D modeling workflows, pose generation, or docking-based comparison studies.
• Strong scripting and automation skills, preferably in Python.
• Ability to design reproducible model workflows with attention to runtime and scalability.
• Strong collaboration and communication skills for working across scientific and technical teams.

Preferred Qualifications

• Experience supporting drug discovery or lead optimization projects.
• Familiarity with co-folder models, Boltz-2, AISB, DeepAutoQSAR, GatorAffinity, MPNN, MM-GBSA, or FEP-related methods.
• Experience designing automated scientific workflows from sequence input to structure and affinity output.
• Exposure to building model benchmarks using project-specific thresholds and precision/recall metrics.
• Ability to work independently and deliver high-quality technical solutions in a contractor setting.

5+ years of experience.