De trochbraak yn eiwitvâldzjen en wat it werklik mooglik makket
December 14, 2025 · Frisian News
DeepMind's protein folding models have solved a decades-old puzzle, but the real-world applications remain limited and expensive. Drug companies still need to prove these tools cut development costs and timelines.
FryskEin 2020 kundige DeepMind oan dat it foarsizze koe hoe't eiwitten har yn trije dimensjes fâldzje mei ferbjusterjende krektens. De trochbraak like in nij tiidrek yn genêskunde en biology yn te ynliede. Dochs wurkje de measte medisijnmakkers fiif jier letter noch altyd op de âlde manier. Eiwitvâldzjen lost ien probleem op. It liedt net automatysk ta wurkjende medisinen of goedkeaper ûndersyk. De masines kenne de foarm no. It witten fan de einfoarm fan in eiwit fertelt wittenskippers wat it dwaan kin, mar net hoe't se dy ynformaasje brûke moatte om sykte te behanneljen. In farmaseutysk bedriuw moat noch altyd miljoenen ferbinings screene, talleaze tests útfiere en him troch de regeljouwing hinne wurkje. De kosten foar it bringen fan in medisyn fan it lab nei de apotheek binne amper feroare. De measte skattings pleatse medisijnûndersyk op 1 oant 3 miljard dollar per goedkard medisyn, ûnferoare fan tsien jier lyn.
Guon bedriuwen hawwe ark om dizze fâldzjingsmodellen hinne boud. Se brûke AI om foar te sizzen hokker medisijnkandidaten wurkje sille, hokker mislearje moatte en hokker lab-ûndersiken rjochtfeardigje. Dizze wurkwize spaart yndie tiid yn iere stadia. Mar it werklike knyppunt docht him letter foar, yn bistûndersiken en minsklike stúdzjes, dêr't regeljouwing en biology alles ferlangsamje. Gjin algoritme foarseit hoe't in minsklik lichem op in nije stof reagearret. Gjin kompjûter foarkomt in mislearre proef.
De echte winners oant no ta binne akademyske ûndersikers en lytsere biotechbedriuwen dy't gjin grutte budzjetten hawwe. Se kinne no eiwitstruktuer bestudearje sûnder miljoenen út te jaan oan laboratoariumapparatuer. Universiteiten yn earme lannen kinne no sykten modelleare dy't eartiids ûnberikber wiene. Dat is wichtich, ek al fernijt it de medisynindustry net fan de iene dei op de oare. Kennis ferspriedt him sneller as ark goedkeap wurde.
Ynvestearders en techbedriuwen hawwe eiwitvâldzjen as game changer promoate. It wie in game changer, mar om oare redenen as advertearre. It ark wurket. De tapassingen yn genêskunde bliuwe beskieden en lizze noch jierren fuort. Elkenien dy't bewearet dat dizze modellen medisijnkosten al ferleage of ynnovaasje ferdubele hawwe, ferkeapet hype, gjin wittenskip.
In late 2020, DeepMind announced it could predict how proteins fold in three dimensions with stunning accuracy. The breakthrough seemed to promise a new era in medicine and biology. Yet five years later, most drug developers still work the old way. Protein folding solves one problem. It does not automatically turn that knowledge into working drugs or cheaper research.
The machines know the shape now. Knowing a protein's final form tells scientists what it might do, but not how to use that information to treat disease. A pharmaceutical company still needs to screen millions of compounds, run countless tests, and navigate regulatory approval. The cost of bringing a drug from lab to pharmacy has barely budged. Most estimates place drug development at 1 to 3 billion dollars per approved medicine, unchanged from a decade ago.
Some companies have built tools around these folding models. They use AI to predict which drug candidates will work, which should fail, and which warrant lab testing. This workflow does save time in early stages. But the actual bottleneck happens later, in animal trials and human studies, where regulation and biology slow everything down. No algorithm predicts how a human body will react to a new compound. No computer avoids a failed trial.
The real winners so far have been academic researchers and smaller biotech firms that lack big budgets. They can now study protein structure without spending millions on lab equipment. Universities in poor countries can now model diseases once out of reach. That matters, even if it does not revolutionize the drug industry overnight. Knowledge spreads faster when tools become cheap.
Investors and tech firms hyped protein folding as a game changer. It was a game changer, but for different reasons than advertised. The tool works. The applications in medicine remain modest and years away. Anyone claiming these models have already cut drug costs or doubled innovation speed is selling hype, not science.
Published December 14, 2025 · Frisian News · Ljouwert, Fryslân