AI drug design & AlphaFold applications

01Overview and value chain

Markers: [EC: FDA AI/ML in Drug Development + EMA AI reflection paper | OECD: Bio-pharmaceuticals | Regulator: FDA (USA), EMA (EU), NMPA (China)]

AI drug design moves hit and lead finding from wet-lab trial-and-error to in silico prediction. Historically a single therapy took roughly a decade and several billion dollars to develop, with about 90% of candidates failing in clinical trials on efficacy or toxicity; generative AI and deep learning now compress the discovery phase from years to months and cut early-stage cost by an estimated 70-80%. Three pillars define the field: AlphaFold-class structure prediction, which yields atomic-accuracy 3D models of proteins and, with AlphaFold 3, of protein complexes with RNA, DNA and ligands; generative chemistry, where GAN, VAE and diffusion models design de novo small molecules and antibodies optimized for binding affinity, solubility, low toxicity and synthetic accessibility; and virtual screening at billion-compound scale (Enamine REAL-class libraries) ranked by physics-based free-energy perturbation (FEP). A lab-in-the-loop of robotic wet-lab feedback retrains the models. The pipeline is regulated end-to-end by the FDA, EMA and NMPA, under frameworks such as the FDA’s AI/ML in Drug Development guidance and the EMA AI reflection paper. The six organizations in this projection span structure prediction (Isomorphic Labs, EMBL-EBI), generative platforms (Insilico Medicine, Exscientia, Recursion) and physics-based modeling (Schrodinger).

Key directions of AI drug design:

1. Structure prediction (AlphaFold 3): atomic-accuracy protein and protein-ligand/RNA/DNA complex structures, enabling de novo target modeling.
2. Generative chemistry: GAN and diffusion models generate de novo molecules tuned for affinity, ADMET profile and synthetic accessibility.
3. Billion-scale virtual screening + FEP: AI screens ultra-large make-on- demand libraries and physics-based FEP ranks leads for binding affinity.
4. Lab-in-the-loop / active learning: robotic phenotypic screening feeds measured biology back to retrain the models.

Sectoral value chain

[target discovery] ──> [generative design] ──> [virtual screening] ──> [CRO synthesis]
                                                              │
                                                  (measured biology feedback)
                                                              │
                                                              ▼
[IND & trials] <─── [in vitro validation] <─────┘

Value chain levels

Cross-cutting technologies of the sector:

• alphafold: deep-learning 3D structure prediction (AlphaFold 3 diffusion module).
• generative-chemistry: GAN/VAE/diffusion de novo molecule generation (Chemistry42).
• free-energy-perturbation: physics-based binding-affinity ranking (FEP+).

02US

The United States leads AI-pharma venture funding, foundation-model development and clinical translation of AI-designed molecules.

AlphaFold alliances, phenotypic screening, physics-based modeling, FDA

• Isomorphic Labs: the Google DeepMind spin-out commercializing AlphaFold 3; in May 2026 it expanded drug-discovery alliances with Eli Lilly and Novartis valued at around USD 3 billion.
• Recursion Pharmaceuticals: operates one of the largest robotic phenotypic-screening facilities for lab-in-the-loop data generation (Recursion OS); it combined with Exscientia in November 2024 to form an end-to-end AI discovery company.
• Schrodinger: the standard for physics-based FEP+ software used across big pharma for binding-affinity prediction and lead optimization.
• FDA CDER AI/ML in Drug Development: the regulatory framework under which first-in-human trials of fully AI-generated molecules are being cleared.

03CN

China is rapidly scaling AI-pharma infrastructure around Shanghai and Beijing research hubs, with IT giants entering biocompute and a landmark AI-originated clinical asset.

generative AI pharma, IT entrants, NMPA guidance

• Insilico Medicine (Shanghai R&D): the Pharma.AI / Chemistry42 platform; its generative-AI-discovered TNIK inhibitor ISM001-055 (rentosertib) for idiopathic pulmonary fibrosis reported positive Phase IIa topline results (published in Nature Medicine) and is advancing toward pivotal trials — a flagship AI-designed asset.
• Tencent and Baidu: Tencent AI Lab’s FineRx (protein-structure prediction and virtual screening) and Baidu’s PaddleHelix biocompute platform.
• NMPA AI guidance and Tsinghua hubs: national guidance on AI/ML in drug development and university accelerators for de novo antibody design.

04EU

The European Union leads fundamental structural bioinformatics, the open AlphaFold structure database, and Oxford-anchored generative discovery.

AlphaFold DB, Oxford discovery pioneer, EU funding

• EMBL-EBI (Hinxton, UK): hosts the AlphaFold Protein Structure Database with Google DeepMind, providing structure coverage for over 214 million protein sequences (open access); its DeepMind partnership was renewed and high-confidence heterodimer predictions were added in 2026.
• Exscientia (Oxford, UK): the Centaur Chemist platform; in 2020 it became the first company to advance an AI-designed drug (DSP-1181) into human clinical trials, and since November 2024 it has operated as part of Recursion.
• BenevolentAI and Horizon Europe: predictive target-discovery platforms for rare diseases, supported by EU AI-in-healthcare funding programmes.

05Leading companies and research institutes

06Tech stack and innovations

The stack combines structure prediction, generative chemistry and physics- based thermodynamics, closed by a lab-in-the-loop.

1. AlphaFold 3 diffusion architecture:
   • a Pairformer module encodes the input sequences and builds a pairwise representation of atomic relationships, then a diffusion module refines a random atomic cloud into the final 3D coordinates over successive denoising steps.
   • predicts protein complexes with RNA, DNA and ligands at near-experimental accuracy, enabling de novo target modeling (Isomorphic Labs).
2. Generative chemistry with reinforcement learning:
   • a generator (GAN or diffusion) proposes a molecule as a SMILES string while AI critic models score it on synthetic accessibility, solubility, ADMET properties and mutagenicity (Ames test).
   • the reward loop iterates millions of times until the molecule meets the design specification (Insilico’s Chemistry42).
3. Physics-based free-energy perturbation (FEP):
   • molecular-dynamics thermodynamic cycles compute the change in binding free energy from a chemical edit (e.g., swapping one functional group), predicting whether it improves binding at roughly 1 kcal/mol accuracy.
   • Schrodinger’s FEP+ replaces expensive preparative synthesis in big-pharma lead optimization.

07Value chains and production pipelines

Industrial pipeline of an AI-designed kinase inhibitor (FDA/IND-enabling)

┌───────────────────────────┐      ┌───────────────────────────┐
│ 1. Target structure       │ ───> │ 2. Generative design      │
└───────────────────────────┘      └───────────────────────────┘
                                                 │
                                                 ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 4. CRO synthesis          │ <─── │ 3. ADMET + FEP hit pick   │
└───────────────────────────┘      └───────────────────────────┘
              │
              ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 5. SPR IC50 validation    │ ───> │ 6. IND filing & Phase I   │
└───────────────────────────┘      └───────────────────────────┘

Stage 1: Target structure prediction

The disease target’s sequence is submitted to AlphaFold 3 to generate an atomic 3D model of the binding pocket, imported into the discovery LIMS.

Stage 2: Generative design

A generative chemistry engine (e.g., Chemistry42) designs a library of candidate molecules complementary to the pocket shape, blocking substrate binding.

Stage 3: ADMET and FEP hit selection

Predictive models filter the library for solubility, hepatocyte stability and selectivity, and FEP ranks the top molecules by computed binding affinity to a short hit list.

Stage 4: CRO synthesis

The selected hit structures (SMILES) are sent to a contract chemistry organization for de novo synthesis, with purity verified by LC-MS.

Stage 5: SPR IC50 validation

Candidates are screened by surface plasmon resonance (SPR) against the immobilized target to measure binding kinetics and rank a confirmed lead by IC50.

Stage 6: IND filing and Phase I

The lead is scaled under cGMP, preclinical safety is established, and an IND dossier is filed with the FDA to begin first-in-human Phase I trials.