School of Bioinformatics · AI & machine learning
Deep Learning for Genomics & Drug Discovery
Deep learning now powers sequence models, protein language models and molecular property prediction. This course teaches how these models work and how to build and evaluate them responsibly on biological data.
Course overview
Deep Learning for Genomics & Drug Discovery begins with the essentials of neural networks in PyTorch and moves quickly to biology: convolutional networks that read DNA, embeddings from protein language models, and graph neural networks for molecules.
Every model is compared with a simple baseline and evaluated with splits that prevent leakage between similar sequences or molecules. You finish with a model card that states what the model is for and where it should not be trusted.
Learning outcomes
By the end of the course you will be able to:
- Build and train neural networks in PyTorch
- Train convolutional models on DNA sequences and interpret them
- Use protein language-model embeddings for prediction tasks
- Build graph neural networks for molecular properties
- Evaluate models with leakage-free splits and write model cards
Skills you will practise
Eligibility and prerequisites
Who can apply
- Students and graduates of B.Sc., M.Sc., B.Tech., M.Tech., B.Pharm., M.Pharm., MBBS, BDS, B.V.Sc. or a PhD programme in the life sciences, biotechnology, bioinformatics, pharmacy, medicine, agriculture, chemistry, statistics or computer science
- Researchers, faculty and industry professionals moving into data-driven biology
- Final-year students may apply before their results are declared
Prerequisites
- Machine Learning for Biology or equivalent
- Python with NumPy and pandas
Best suited to
- Students who have completed Machine Learning for Biology
- Computational biologists adopting deep learning
- Engineers and data scientists moving into biotech
Module-wise syllabus
6 modules. Each pairs instructor-led sessions with a hands-on lab on public data.
Topics
- Tensors and autograd
- Training loops and regularisation
- GPU training on free notebooks
Topics
- One-hot encoding
- Convolutional networks for motifs
- Attribution maps
Topics
- Embeddings from ESM-style models
- Fine-tuning versus frozen features
- Homology-aware splits
Topics
- Molecules as graphs
- Message passing
- Scaffold splits
Topics
- Baselines and ablations
- Uncertainty and applicability
- Model cards
Topics
- Problem framing
- Model and baseline
- Model card
Practical sessions and projects
Capstone project
Build a model you can defend
Train a deep-learning model on a biological problem, beat a strong baseline under a leakage-free split, and document it with a model card.
Assessed on the correctness of the analysis, the quality of the code and the reasoning in the write-up. A pass is required for the certificate.
- Neural networks in PyTorch
Build and train a small network, then diagnose over-fitting with learning curves.
- Models that read DNA
Train a CNN to predict transcription-factor binding and recover the motif from its filters.
- Protein language models
Predict a protein property from embeddings and show how a random split inflates accuracy.
- Graph neural networks for molecules
Train a GNN on a solubility dataset and compare it with a fingerprint-based random forest.
- Responsible evaluation
Write a model card for your best model, including failure cases.
- Project: a defensible model
Deliver a deep-learning model on a biological dataset with a baseline, leakage-free evaluation and a model card.
Software, tools and datasets
Software and tools
All software is free and open source or free for academic use. Recommended computer: 8 GB RAM, 30 GB free disk, Linux, macOS or Windows with WSL2.
Datasets
- ENCODE transcription-factor binding data: Sequence models.ENCODE (public)
- MoleculeNet benchmarks: Molecular property prediction.moleculenet.org
Core reading
- Eraslan G, Avsec Ž, Gagneur J, Theis FJ (2019). Deep learning: new computational modelling techniques for genomics. Nature Reviews Genetics 20:389–403.
- Lin Z et al. (2023). Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 379:1123–1130.
Duration, format and fees
- Next cohort
- Dates are sent with your admission offer
- Timetable
- Three evening or weekend sessions a week (IST), plus a lab clinic; intensive and weekend-only schedules on request
- Payment
- Secure online payment link with your admission offer (UPI, cards, net banking). Instalments on request for fees above ₹15,000.
- Refunds
- See the refund and cancellation policy.
Instructor
The instructor for each cohort is named in your admission offer and on this page when the cohort is announced. Vedicgen instructors are selected for hands-on research or industry experience in the subject they teach, and every course is reviewed by a second subject expert.
Questions about who teaches a cohort? [email protected]
Certificate
Learners who complete the course and pass its capstone receive a Vedicgen certificate of completion for Deep Learning for Genomics & Drug Discovery, issued by Vivofold, stating the guided hours and skills assessed. Each certificate has a unique ID and QR code that anyone can check.
Apply for Deep Learning for Genomics & Drug Discovery
Apply
Send the form below. It takes about five minutes.
Review
An adviser confirms your fit and format within two working days.
Offer
You receive dates, timetable and a secure payment link.
Onboard
Set-up guide, pre-reading and joining links.
Frequently asked questions
No. Labs run on free GPU notebooks (for example Google Colab or Kaggle). A laptop with a modern browser is enough.
In the live cohort, modules are taught in instructor-led sessions with a weekly lab clinic where you work on real public data with help at hand; every session is recorded. The self-paced format uses the same lessons, labs and datasets with mentor review of your capstone. Both formats end with an assessed project reviewed for the code and the science.
A laptop with at least 8 GB of RAM and about 30 GB of free disk space, running Linux, macOS, or Windows with WSL2. Installation instructions are sent before the first session, and lab clinics help with any set-up problems.
Yes. Learners who complete the course and pass its assessment receive a Vedicgen certificate of completion, issued by Vivofold. Every certificate carries a unique ID and a QR code that anyone, including an employer or admissions office, can check on the certificate verification page.
Apply with the form on this page. Live and in-person cohorts run through the year; the dates, timetable and joining details of the next cohort are sent with your admission offer and listed on this page once announced. Self-paced enrolment opens as soon as your application is accepted.
Yes. No course runs longer than four weeks, and the length is customisable: choose the standard schedule, an intensive one- to two-week schedule, weekends only, or a custom timetable. Pick your preference on the application form; colleges and teams can book their own dates.
Part of a programme
Take this course as part of a programme and save.
Continue learning