AlphaFold

AlphaFold is an artificial intelligence system developed by Google DeepMind that predicts the three-dimensional structure of proteins from their amino acid sequence. The protein folding problem — determining how a linear chain of amino acids folds into the precise three-dimensional shape that determines its biological function — had been one of the central unsolved challenges in biology for over fifty years. AlphaFold2, released in 2020, effectively solved this problem for single-chain proteins, achieving accuracy that rivals experimental methods such as X-ray crystallography and cryo-electron microscopy.

The system achieved global recognition at CASP14 (the 14th Critical Assessment of Protein Structure Prediction competition) in December 2020, where AlphaFold2 attained a median GDT score above 92 — a level previously considered unattainable for AI methods and within the margin of experimental error. In October 2024, Demis Hassabis (CEO of Google DeepMind) and John Jumper (the lead scientist of the AlphaFold project) were awarded the Nobel Prize in Chemistry for this achievement.

Technical Architecture

AlphaFold2

AlphaFold2 uses a neural network architecture called Evoformer, a modified transformer that operates on two types of representations simultaneously: a multiple sequence alignment (MSA) representation capturing evolutionary variation across related proteins, and a pair representation encoding relationships between residue pairs.

The MSA representation encodes information from hundreds or thousands of related protein sequences across species. Evolutionary pressures tend to preserve co-evolving residue pairs that are spatially close in the folded structure, so correlated variation across species provides geometric constraints. Evoformer iteratively updates both representations through a series of attention operations that allow information to flow between the MSA and pair representations.

The final structure prediction is produced by a structure module that places amino acid backbone atoms (N, CA, C, O) in three-dimensional space using equivariant representations that respect the symmetries of 3D space. The network outputs atomic coordinates along with per-residue confidence scores (pLDDT) and predicted alignment errors (PAE) that indicate which parts of the structure are reliable.

AlphaFold3

AlphaFold3 was jointly developed by Google DeepMind and Isomorphic Labs and announced in May 2024. It significantly extends the scope of prediction beyond single-chain proteins. AlphaFold3 can predict the structure of complexes involving proteins, DNA, RNA, small molecules (ligands), ions, and post-translational modifications such as glycosylation and phosphorylation.

The architecture replaces the Evoformer with a simplified Pairformer and incorporates a diffusion module — borrowing from image generation models — that iteratively refines atomic positions starting from a noisy cloud of atoms. This approach handles the diverse molecular types more naturally than the residue-centric Evoformer. AlphaFold3 achieves state-of-the-art performance on protein-ligand docking and protein-nucleic acid structure prediction benchmarks.

Access to AlphaFold3 is provided through the AlphaFold Server, a free web interface for non-commercial research use. Commercial applications require a licence from Isomorphic Labs.

Impact on Structural Biology

Prior to AlphaFold, the Protein Data Bank (PDB) contained approximately 170,000 experimentally determined structures after more than fifty years of labour-intensive experimental work. AlphaFold2 produced high-confidence predictions for over 200 million proteins — essentially the entire known proteome of life — which are freely available through the AlphaFold Protein Structure Database maintained jointly by DeepMind and EMBL-EBI. By November 2025, the system was in use by over 3 million researchers across 190 countries.

Structural biology has been accelerated in several areas. Researchers studying neglected tropical diseases — conditions that disproportionately affect low-income countries and historically attract little pharmaceutical investment — now have structural data for pathogen proteins that would have required years and millions of dollars to obtain experimentally. Vaccine developers have used AlphaFold structures to identify surface-exposed epitopes for antigen design. Enzyme engineers have used predicted structures to guide directed evolution campaigns.

Drug Discovery Applications

Drug discovery is the most commercially significant application of AlphaFold. Understanding the three-dimensional structure of a disease-relevant protein is a prerequisite for structure-based drug design, where medicinal chemists use the binding pocket geometry to rationally design small molecules that inhibit or activate the protein.

Isomorphic Labs, a subsidiary of Alphabet, was established specifically to apply AlphaFold and related AI methods to drug discovery. By 2024, Isomorphic had announced partnerships with Eli Lilly and Novartis to apply its platform to their drug discovery pipelines. Several biotech companies including Recursion Pharmaceuticals, Exscientia, and Insilico Medicine use AlphaFold structures as inputs to their AI drug discovery workflows.

AlphaFold3's ability to predict protein-ligand binding poses makes it directly applicable to virtual screening — computationally testing millions of candidate molecules for binding before conducting experiments. Early benchmarks showed AlphaFold3 outperforming established docking programs such as Glide and AutoDock Vina on standard benchmarks.

Malaysian research institutions have adopted AlphaFold to accelerate biomedical and pharmaceutical research, particularly in areas of regional disease burden. The Institute for Medical Research (IMR) under the Ministry of Health Malaysia has used AlphaFold to study proteins from pathogens relevant to Malaysia including the dengue virus, Plasmodium falciparum (malaria), Burkholderia pseudomallei (melioidosis), and Nipah virus. Melioidosis, caused by a soil bacterium endemic to northern Malaysia and much of Southeast Asia, has been a focus of structure-guided vaccine research using AlphaFold-predicted antigen structures.

Universiti Malaya (UM), Universiti Kebangsaan Malaysia (UKM), and Universiti Sains Malaysia (USM) have integrated AlphaFold into their structural bioinformatics and drug discovery research programs. USM's School of Pharmaceutical Sciences has published work applying AlphaFold structures to natural product drug discovery, leveraging Malaysia's rich biodiversity of medicinal plants.

The Malaysian pharmaceutical sector, while primarily focused on generic manufacturing rather than drug discovery, is beginning to invest in computational drug design capabilities. Companies such as Hovid Berhad and CCM Pharmaceuticals have engaged with Malaysian universities to explore AI-guided formulation and process optimisation. The Malaysian Bioeconomy Development Corporation (Bioeconomy Corporation) has identified AI-accelerated drug discovery as a strategic sector, and AlphaFold competence is increasingly mentioned in job postings from pharmaceutical companies operating in Malaysia.

MDEC's Digital Innovation Hub programme has supported computational biology startups leveraging AlphaFold, recognising structural AI as a key technology for Malaysia's ambition to grow its life sciences and biotech export sector. The government's push to develop Malaysia as an ASEAN biomedical hub, anchored at Selangor's Bio-XCell integrated life sciences park, includes investment in digital tools for drug development in which AI protein modelling plays an increasing role.

See Also

1. Jumper, J., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583-589.
2. Abramson, J., et al. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 630, 493-500.
3. The Nobel Prize in Chemistry 2024. (2024). Nobel Prize Announcement. nobelprize.org.
4. Varadi, M., et al. (2022). AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Research, 50(D1), D439-D444.
5. Google DeepMind. (2024). AlphaFold Server launch. deepmind.google/science/alphafold.