Vivofold Research
AI & Machine Learning for Biology
Machine learning in biology often fails quietly through data leakage, weak baselines and unstated limits. We study and build models that hold up: evaluated with splits that respect biology, compared with simple baselines, and documented with model cards.
Questions
What we want to know
- When do large models beat simple baselines on biological tasks, once leakage is removed?
- How should applicability domains be communicated to non-specialist users?
- How accurate are language models at extracting structured data from biological literature?
Approach
How we work on it
- Homology-, cluster- and scaffold-aware evaluation
- Baseline-first model development
- Model cards and uncertainty reporting
- Measured evaluation of LLM tools against labelled data
Methods already implemented
Cluster cross-validation and Y-randomisationApplicability-domain estimationNMF signature extraction
Open projects
Projects you can join
01
Leakage in biological ML benchmarks
Re-evaluate public benchmarks with biologically aware splits.
02
Literature extraction accuracy
Measure how well LLM pipelines extract experimental parameters from methods sections.
03
Communicating uncertainty
Design and test ways of showing model applicability to bench scientists.
Across Vivofold
Learn it, use it, commission it
Vedicgen courses
Machine Learning for BiologyUp to 4 weeks, about 32 hours live (customisable)Dry labfrom ₹5,999 + GSTDeep Learning for Genomics & Drug DiscoveryUp to 4 weeks, about 32 hours live (customisable)Dry labfrom ₹5,999 + GSTLarge Language Models for Life SciencesUp to 4 weeks, about 32 hours live (customisable)Dry labfrom ₹4,499 + GST
AI & Machine Learning in Biology Internship Collaborate