AI in Biology [2026]: 7 Breakthroughs Changing Medicine Right Now

DeepMind's AlphaFold 2 (2020) was described at the time as one of the most significant scientific achievements of the decade — solving the protein folding problem that structural biologists had worked on for fifty years. Here is the honest assessment of what AlphaFold actually changed, what it didn't change, and what AI in biology looks like in 2026.

What AlphaFold Actually Did

The protein folding problem — predicting a protein's three-dimensional structure from its amino acid sequence — was a 50-year challenge in structural biology. A protein's structure determines its function, and knowing the structure is essential for understanding how proteins work and for designing drugs that interact with them. Before AlphaFold 2, determining a protein's structure experimentally (using X-ray crystallography, cryo-electron microscopy, or NMR) was expensive, time-consuming, and not possible for every protein. AlphaFold 2 predicted protein structures with accuracy comparable to experimental methods for a large fraction of proteins — a genuine breakthrough in speed and accessibility of structural information.

AlphaFold's database, released publicly in 2021 and expanded subsequently, provided predicted structures for essentially the entire proteome of model organisms including humans. This made structural information available for proteins that had never been experimentally determined, at no cost and immediately. The scientific community's uptake has been substantial — structural biologists have incorporated AlphaFold predictions into research that would previously have required years of experimental structural determination.

What AlphaFold Didn't Change

The "solved protein folding" framing overstated the achievement in specific ways. AlphaFold excels at predicting the structure of stable, well-folded proteins but has limitations with intrinsically disordered regions (protein segments that don't adopt a fixed structure), with protein-protein interactions in context (how a protein's structure changes when bound to other proteins), and with proteins whose structure changes dynamically rather than being fixed. These limitations matter specifically for drug discovery — many drug targets involve protein-protein interactions and conformational changes that AlphaFold's static structure predictions don't fully address.

Drug discovery has been assisted but not transformed on the timelines that early enthusiasm projected. The drug development pipeline from structure to clinical candidate to approved drug involves many steps beyond structural prediction, and AlphaFold has accelerated certain early stages without compressing the overall timeline dramatically. The drugs enabled by AlphaFold are in early stages of development; the clinical validation of AlphaFold-assisted drug discovery as a meaningfully better approach than previous methods is still being established.

What Comes Next in AI Biology

The 2024-2026 period has seen significant extension of AI methods in biology beyond protein structure: protein design (using AI to design novel proteins with specified properties, which is distinct from predicting existing protein structures), genomics (predicting regulatory effects of genetic variants), and drug-molecule property prediction. These applications are earlier in development than protein structure prediction but represent the growing frontier of AI-biology integration that's producing real tools alongside the hype.

My honest take: AlphaFold was a genuine breakthrough that made structural information accessible at unprecedented scale. The "protein folding solved" framing overstated it — disordered regions and conformational dynamics remain hard. Drug discovery benefit is real but not yet as dramatic as early projections. The next frontier is protein design, not just structure prediction.