Artificial Intelligence and Large Language Models for Biology: Generative and Reproducible Design for Biological Models

From "what is a FASTQ file" to models you can defend.

A self-contained course that takes you from the first line of shell to building and honestly evaluating deep-learning and generative models on genomic, single-cell, spatial, and histopathology data — and applying them to disease biology and drug discovery. Every concept is explained from first principles, with runnable code, comparison tables, and worked exercises. Nothing here says "see the literature" in place of an explanation.

Written for three readers at once

• Wet-lab biologists and clinicians who know the biology and the data, and need the command line, the statistics, and the modelling.
• CS and ML people who know models and evaluation, and need molecular biology, assay artefacts, and the reasons biological data breaks i.i.d. assumptions.
• Bioinformaticians levelling up from pipelines and standard analyses into modern single-cell and spatial methods, deep learning, foundation models, and drug discovery.

There is no prerequisite beyond high-school biology and a willingness to type commands into a terminal.

In this volume

• Module 11 — LLMs, and Language Models of Biology. Tokenisation through the KV cache, RLHF and DPO, prompting and structured output, LoRA and QLoRA with a full training script, a complete PubMed retrieval system with citation verification, agents, hallucination and prompt injection, and ESM, Evo and scGPT read as language models.
• Module 12 — Generative AI and Generative Design. VAEs, GANs, normalising flows, diffusion and flow matching with the objectives written out; molecular generation and multi-parameter optimisation; RFdiffusion and ProteinMPNN design-and-filter pipelines; perturbation prediction; honest evaluation; biosecurity.
• Module 13 — Computational Drug Discovery and Development. Pipeline economics and attrition, genetics-led target validation with a reusable scorecard, RDKit cheminformatics, docking and free-energy methods assessed honestly, QSAR and ADMET, structure-based design after AlphaFold, biologics, repurposing, trials and regulation, and a worked go/no-go project.
• Module 14 — Disease Biology, Human Genetics, Precision Medicine. GWAS and fine-mapping, colocalisation, Mendelian randomisation, polygenic scores and portability, ACMG variant classification, cancer drivers and mutational signatures, liquid biopsy statistics, neoantigens, pharmacogenomics, multi-omic survival modelling.
• Module 15 — Reproducible Practice and Ethics. Environments and containers, workflow managers, data management, provenance, and the accountability that computational claims about patients require.
• Module 16 — Capstones and Roadmap. Ten graded projects with named public datasets, deliverables, traps, and rubrics.
• Glossary and cheatsheets for the whole course.

The companion volume, Bioinformatics and Multi-Omics, covers the working environment, molecular biology as a data model, sequences and databases, sequencing end to end, bulk and single-cell and spatial transcriptomics, digital pathology, and the machine-learning and deep-learning foundations.

Dr. Ashok Kumar Dwivedi