Intended Audience

Computational biologists, bioinformaticians, and data scientists in biopharma or academic research.

Familiarity with Python, R, or command-line scripting is recommended. No prior experience with LLM coding tools is required.

Problem Statement

Coding assistants and agentic workflows are widely used in software engineering, but general software tooling does not cleanly translate to computational biology. Biological data is noisy, formats are idiosyncratic, and subtle errors in analysis logic can go unnoticed without domain-specific validation.

This workshop introduces “Scientific Specification-Driven Development”—adapting software spec design to computational biology. Participants write structured specifications, use models to generate analysis code, and establish validation loops that evaluate outputs directly against biological ground truth and expected data distributions.

Syllabus & AI Competencies Covered

• LLM coding workflows: Setting up command-line and IDE-based coding agents for data analysis.
• Specification-driven development: Writing clear analysis specifications and using LLM critique cycles before generating code.
• Context and tool integration: Providing models with domain schemas, reference data, and external tools via Model Context Protocol (MCP).
• Automated testing on biological data: Designing validation loops, sanity checks, and unit tests to verify scientific outputs.
• Refactoring and debugging: Directing models to debug complex bioinformatics scripts and adapt code to new datasets.

Example Work Product

An interactive single-cell RNA-seq analysis dashboard.

During the workshop, participants use a foundation model to process single-cell genomics data and answer an exploratory research question. Participants build custom tool integrations to query public datasets, validate cell type predictions, and assemble an interactive dashboard summarizing the results.

Sample final projects are available on the class website.

Contact