Training / Affinity
Affinity, the computational chemist training
Proteins move. Ligands bind. Learn to predict both.
Affinity is a three-month, research-grade training in protein–ligand and protein–binder modelling, for life-science graduates who want the drug discovery roles that hire for it.
- Starts
- January
- Length
- 3 months, 5–6 hours a week
- Selection
- Two rounds, about two days
- Fees
- Paid monthly
Pick the role. The training bends around it.
Everyone completes the core curriculum: all seven modules. Your target role sets your specialisation: the techniques you take furthest, your project, and the openings we push you towards.
Structure-Based Design
- Computational Medicinal ChemistApply for this role
- Structure-Based Drug Design ScientistApply for this role
- Virtual Screening and Docking SpecialistApply for this role
- Cheminformatics ScientistApply for this role
You go furthest in
- Fold
- Lead
A project might be
Screen and triage a ligand library against a kinase target, then defend your top twenty.
Who hires for it
Pharma and biotech R&D · Research services (CRO and CRDMO)
Simulation and Free Energy
- Biomolecular Simulation (MD) ScientistApply for this role
- Free-Energy (FEP) ScientistApply for this role
- QM/MM and Quantum Chemistry ScientistApply for this role
You go furthest in
- Motion
- Orbital
A project might be
Rank a congeneric series by relative binding free energy and compare it with experiment.
Who hires for it
Pharma and biotech R&D · Scientific software
AI-Driven Discovery
- AI/ML Scientist, Drug DiscoveryApply for this role
- Generative Chemistry ScientistApply for this role
- Machine-Learned Potentials ScientistApply for this role
You go furthest in
- Binder
- Lead
A project might be
Train a leakage-safe affinity model, then use it to steer a generative design loop.
Who hires for it
AI-first drug discovery · Pharma and biotech R&D
Biologics and Protein Design
- Protein and Binder Design ScientistApply for this role
- Protein Language Model ScientistApply for this role
You go furthest in
- Binder
- Fold
- Motion
A project might be
Design and rank protein binders against a target interface, then test their stability in simulation.
Who hires for it
AI-first drug discovery · Pharma and biotech R&D
New Modalities
- Covalent Drug Design ScientistApply for this role
- Allosteric and Protein–Protein Interaction ScientistApply for this role
- Cryptic Pocket Discovery ScientistApply for this role
You go furthest in
- Motion
- Fold
- Orbital
A project might be
Reveal a cryptic pocket with enhanced sampling, then dock into it.
Who hires for it
Pharma and biotech R&D · Research services
Scientific Platforms
- Research Software EngineerApply for this role
- Pipeline and Workflow DeveloperApply for this role
- Research Computing (HPC/GPU) EngineerApply for this role
You go furthest in
- Backbone
- Motion
A project might be
Build a reproducible docking-to-dynamics pipeline that runs on a GPU cluster.
Who hires for it
Scientific software · AI-first drug discovery
Seven modules, named the way a protein is built.
Core curriculum: all seven modules, for everyone. Specialisation: the modules your target role needs, taken furthest and highlighted in orange. The full syllabus is shared with shortlisted applicants.
1PrimaryThe chemistry and biology that never stop mattering
Hybridisation, resonance and conjugation · pKa, protonation and tautomerism · stereochemistry and conformational analysis · protein folding, stability and allostery · Michaelis–Menten kinetics and inhibition modes · ΔG, Kd and Boltzmann statistics · electrostatics, solvation and the hydrophobic effect · regression, error propagation and Monte Carlo sampling
- Paper, then Python
2BackboneThe computer you'll actually work on
Linux processes and permissions · Bash, grep, sed and awk · SSH and rsync to remote clusters · SLURM partitions, job arrays and dependencies · Conda, Docker and Apptainer · NumPy, pandas, RDKit and Biopython · parsing PDB, mmCIF, SDF and MOL2 · Git, testing, logging and provenance · Snakemake and Nextflow · FAIR data
- Bash
- Python
- Git
- Snakemake
- Nextflow
- Docker
- Apptainer
3FoldStructures you can trust, pockets worth targeting
Biological assemblies, resolution, occupancy and B-factors · Ramachandran and clash validation · homology modelling and AlphaFold pLDDT/PAE · protonation, missing atoms and loop rebuilding · cryptic and allosteric pockets · rigid and flexible docking · redocking and cross-docking · enrichment, ROC-AUC and PR-AUC against decoys · consensus scoring and pose clustering
- PyMOL
- ChimeraX
- Foldseek
- AutoDock Vina
- GNINA
- smina
4MotionProteins move, and so should your conclusions
Atom types, charges and ligand topologies · solvation and ionic strength · NVT/NPT equilibration, thermostats and barostats · PME, constraints and hydrogen-mass repartitioning · RMSD, RMSF, Rg, SASA and H-bond occupancy · water networks and pocket dynamics · PCA, clustering and Markov state models · umbrella sampling and metadynamics · MM/GBSA, FEP and thermodynamic integration · lambda schedules and convergence · residence time
- GROMACS
- OpenMM
- AmberTools
- MDAnalysis
- MDTraj
- ProDy
5OrbitalWhen classical mechanics isn't enough
Hartree–Fock and electron correlation · density functional theory and functional choice · basis-set convergence · geometry optimisation, frequencies and thermochemistry · electrostatic potentials and partial charges · implicit and explicit solvation · transition states and reaction barriers · QM/MM at enzyme active sites · electron-density analysis
- Psi4
- xTB
- NWChem
- CP2K
- ORCA
- Avogadro
- Multiwfn
6LeadFrom a hit to something worth making
Target tractability and druggability · pharmacophores and interaction fingerprints · SAR and SPR · bioisosteres, scaffold hopping, fragment growing and linking · ligand efficiency and LipE · logP, logD, PSA and solubility · CYP metabolism and structural alerts · hERG, hepatotoxicity and genotoxicity · PAINS and aggregators · QSAR applicability domains · multi-parameter optimisation · patent and novelty checks
- RDKit
- SwissADME
- pkCSM
- ADMETlab
- Open Babel
7BinderWhere machine learning meets biologics
Leakage-safe splits and nested cross-validation · class imbalance, calibration and uncertainty · graph neural networks and message passing · 3D equivariant networks · protein language models · structure-aware protein–ligand models · diffusion-based design of small molecules and protein binders · learned docking rescoring · machine-learned interatomic potentials · active learning and Bayesian optimisation
- PyTorch
- scikit-learn
- DeepChem
The ground Affinity covers.
Nine stages, 91 techniques, every one practised on real protein systems, on GPUs. The core curriculum takes you through all of it; your specialisation is where you go furthest.
Stage 1 of 9 · Module Fold
Structure
Start from the structure, and learn when not to trust it.
- PDBx/mmCIF and biological assemblies
- Resolution, occupancy and B-factors
- Ramachandran and clash validation
- Homology modelling
- AlphaFold pLDDT and PAE
- Protonation and tautomer states
- Missing-loop reconstruction
- Metals, cofactors, glycans and waters
- Cryptic and allosteric pockets
Stage 2 of 9 · Module Fold
Screening
Search a library fast, then prove the search worked.
- Grid and search-space definition
- Rigid and flexible-receptor docking
- Ligand preparation and stereo enumeration
- Redocking and cross-docking
- CNN rescoring
- Decoy benchmarking
- Enrichment factors
- ROC-AUC and PR-AUC
- Consensus scoring
- Pose clustering
- Covalent and allosteric docking
- Hit triage
Stage 3 of 9 · Module Motion
Dynamics
Let the protein move, and measure what changes.
- Force fields and atom typing
- Ligand parameterisation and partial charges
- Periodic boundaries, solvation and ionic strength
- NVT/NPT equilibration
- Thermostats and barostats
- PME and long-range electrostatics
- Hydrogen-mass repartitioning
- Replicate production runs
- RMSD, RMSF, Rg, SASA and H-bond occupancy
- Water networks
- PCA and clustering
- Markov state models
Stage 4 of 9 · Module Motion
Energetics
Put numbers on binding, with honest error bars.
- MM/GBSA and MM/PBSA
- Alchemical free-energy perturbation
- Thermodynamic integration
- Relative and absolute binding free energies
- Lambda schedules, overlap and convergence
- Umbrella sampling
- Metadynamics
- Residence time and unbinding kinetics
- Uncertainty estimation
Stage 5 of 9 · Module Orbital
Quantum
Go below the force field when electrons matter.
- Born–Oppenheimer approximation
- Hartree–Fock and electron correlation
- Density functional theory
- Basis-set convergence
- Geometry optimisation
- Vibrational frequencies and thermochemistry
- Electrostatic potentials and charges
- Implicit and explicit solvation
- Transition states and barriers
- QM/MM for active sites
- Electron-density analysis
Stage 6 of 9 · Module Lead
Design
Turn a hit into something a chemist would want to make.
- Target tractability
- Pharmacophores and interaction fingerprints
- SAR and SPR
- Bioisosteres and scaffold hopping
- Fragment growing, linking and merging
- Ligand efficiency and LipE
- ADMET and developability
- CYP metabolism and soft spots
- HERG and toxicity alerts
- PAINS and aggregators
- QSAR applicability domains
- Multi-parameter optimisation
Stage 7 of 9 · Module Binder
Learning
Models that generalise, and know when they don't.
- Leakage-safe splits
- Nested cross-validation
- Calibration and uncertainty
- Graph neural networks
- 3D equivariant networks
- Protein language models
- Diffusion-based generative design
- Protein binder design
- Learned docking rescoring
- Machine-learned interatomic potentials
- Active learning
- Bayesian optimisation
Stage 8 of 9 · Module Backbone
Engineering
Make it run again, on someone else's machine.
- Linux and Bash
- SSH and remote clusters
- SLURM job arrays and dependencies
- Conda, Docker and Apptainer
- Snakemake and Nextflow pipelines
- Git and code review
- Tests, logs, checksums and provenance
- GPU resource planning
- FAIR data
Stage 9 of 9 · Every module
Communication
Defend the result to people who know more than you.
- Critical paper reading
- Technical reports
- Publication-quality molecular figures
- Research presentations
- Defending results to researchers
Skills that travel. The Python that ranks ligands also sorts patient cohorts; the statistics that validate a docking run also validate an assay.
Twelve weeks. One real project.
You don't finish with a quiz. You finish with a target, a question, results on cloud GPUs and a talk you give to researchers. Week nine is where it gets interesting.
Each week mixes live sessions, materials for self-preparation, tests with written corrections, open discussion and presentations you give to the batch.
You leave with
- A defended project, published in the AKA Project Archive
- A research talk
- A public portfolio repository
- A CV rewritten for your target role
- A shortlist of openings from Fork
- Applications sent
No certificate-and-wave.
Applications are part of the training. Your mentor reviews your CV, statement of purpose and project write-up, then pushes you towards the right openings: industry roles, research posts and funded PhDs.
Career consulting
One-to-one guidance on where to apply and how, for enrolled students only.
Fork
Our private opportunities board opens to your batch on day one. Look inside Fork
Your mentor: Ashwath Kumar, a computational chemist leading drug discovery programmes. Meet him
Openings.
In chess, the opening decides the game. These are the places that hire for the roles Affinity prepares you for, and where to check what those roles pay.
| Where these roles work | Roles they hire for | Careers pages |
|---|---|---|
| Pharma and biotech R&D | Computational Medicinal Chemist · SBDD Scientist · MD Scientist · Free-Energy (FEP) Scientist | Biocon · Dr. Reddy's · Novartis · AstraZeneca · Eli Lilly |
| Research services (CRO and CRDMO) | Virtual Screening and Docking Specialist · Cheminformatics Scientist · Computational Chemist | Syngene · Jubilant Biosys · Aragen · Sai Life Sciences · Evotec |
| AI-first drug discovery | AI/ML Scientist · Generative Chemistry Scientist · Protein and Binder Design Scientist | Isomorphic Labs · Recursion · Insilico Medicine · Aganitha · Generate:Biomedicines |
| Scientific software and modelling platforms | Free-Energy (FEP) Scientist · Research Software Engineer · Pipeline and Workflow Developer | Schrödinger · Cresset · Certara |
| Academic research | Every specialisation, through funded PhD programmes and research posts | IITs, IISc, CSIR laboratories and universities abroad |
Check pay independently: Glassdoor India and AmbitionBox for India, PayScale for a US benchmark. Search any role title above. We don't quote numbers; they do, from people in the roles.
Names shown to help your research. AKA is not affiliated with these organisations.
Two rounds, about two days.
Here's exactly what happens. This is not for everyone, and that's the point. A small batch means your mentor reads every line you write.
Apply
About ten minutes. Your role, your background, a short statement.
Round one: a short task
Sent by email. It shows us how you think, not what you've memorised.
Round two: a conversation
With the AKA team, about where you want to be in three years.
Your offer
A payment page for the first month arrives with your materials, instructions, class link and syllabus. Fees are collected monthly to keep the load light.
Questions.
Do I need to code?
How do classes run?
Do I get a certificate?
Does it guarantee a job?
What if drug discovery hiring stays slow?
What does it cost?
Who teaches?
Your opening move.
January's batch is small. Round one takes about ten minutes.