🔥 Hot topics · Kan dit NIET · Kan dit · § The Court · Recente omslagen · 📈 Tijdlijn · Vraag · Redactionele stukken · 🔥 Hot topics · Kan dit NIET · Kan dit · § The Court · Recente omslagen · 📈 Tijdlijn · Vraag · Redactionele stukken
Stuff AI CAN'T Do

Kan AI beter presteren dan mensen bij het voorspellen van eiwit-eiwitinteracties ?

Wat denk je?

AlphaFold-Multimer en opvolgers namen deze benchmark in 2024.

Background

Since 2021, deep-learning models have steadily improved PPI prediction by learning co-evolutionary signals and structural constraints from large protein sequence alignments. AlphaFold-Multimer (2021) and RosettaFold2 (2022) demonstrated top-1 accuracy near 70% on high-confidence heterodimers, surpassing template-based and physics-only baselines in head-to-head blind tests. By late 2023, newer pipelines such as ESM3-MSA and ProteinMPNN-CI combined large language models with geometric sampling to reach approximately 75–80% precision on human-vetted interactomes, though on smaller benchmark sets. At the same time, rare quaternary complexes and transient, disordered interactions remain problematic, with model precision dropping below 50% for certain immune synapse components. Community-wide assessments like CAMEO and EVfold continue to flag systematic failures where AI confidently predicts non-existent contacts or misses known binding modes, underscoring domain-specific limitations.

SOURCE: no public reference

Status voor het laatst gecontroleerd op May 15, 2026.

📰

Galerie

In the Court of AI Capability
Summary of Findings
Verdict over time
May 2026May 2026May 2026
Sitting at the Bench Filed · mei 15, 2026
— The Question Before the Court —

Kan AI beter presteren dan mensen bij het voorspellen van eiwit-eiwitinteracties?

★ The Court Finds ★
▲ Upgraded from In_research
Bijna

Er bestaan beperkte demonstraties — maar het panel was niet unaniem.

Ruling of the Bench

The jury found itself swayed by AI’s impressive strides in predicting protein-protein interactions, with most agreeing it has surpassed human performance on curated datasets but still falls short of universal dominance across all biological contexts. Two jurors argued the threshold had been crossed with deep learning models like AlphaFold-Multimer, while the others remained cautious, noting gaps in real-world applicability and the reliance on structural predictions rather than direct experimental evidence. Ruling: "AI knows the dance—now it just needs to lead every step of the ball.

— Hon. G. Hopper, Presiding
Jury Tally
2Ja
3Bijna
0Nee
Verdict Confidence
84%
The Court of AI Capability is, of course, not a real court.
But the data is real.
The Case File · Stacked History
Session I · May 2026 In_research
Session II · May 2026 In_research
Case № 7326 · Session III
In the Court of AI Capability

The Case File

Docket № 7326 · Session III · Vol. III
I. Particulars of the Case
Question put to the courtKan AI beter presteren dan mensen bij het voorspellen van eiwit-eiwitinteracties?
SessionIII (3 hearing)
Convened15 mei 2026
Previously ruledIN_RESEARCH (May '26) → IN_RESEARCH (May '26) → ALMOST (May '26)
Presiding JudgeHon. G. Hopper
II. Cumulative Tally Across Sessions

Across 3 sessions, 11 jurors have heard this case. Combined tally: 6 YES · 3 ALMOST · 2 NO · 0 IN RESEARCH.

Note: cumulative includes older juror opinions. The current session tally above is the live verdict.

III. Verdict

By a vote of 2 — 3 — 0, the panel returns a verdict of BIJNA, with verdict confidence of 84%. The court so orders. Verdict upgraded from prior session.

IV. Verklaringen van het college
Jurylid I ALMOST

"AI models like AlphaFold predict interactions with high accuracy"

Jurylid II ALMOST

"AI outperforms human experts on curated PPI datasets but not universally across all proteins"

Jurylid III JA

"AI models, such as RF2-PPI and AlphaFold-Multimer, have demonstrated high accuracy (up to 90%) in predicting protein-protein interactions, outperforming traditional methods."

Jurylid IV JA

"AlphaFold-Multimer and other deep learning models have demonstrated superior accuracy in predicting protein-protein interactions compared to experimental and traditional computational methods."

Jurylid V ALMOST

"AI models like AlphaFold predict structures, aiding interaction predictions"

Individuele juryverklaringen worden in het oorspronkelijke Engels weergegeven om de bewijsprecisie te behouden.

G. Hopper
Presiding Judge
M. Lovelace
Clerk of the Court

Wat het publiek denkt

Nee 6% · Ja 76% · Misschien 18% 154 votes
Ja · 76%
Misschien · 18%
Trend heeft stemmen van ten minste 2 verschillende dagen nodig.

Discussie

no comments

Opmerkingen en afbeeldingen gaan door een beoordeling door de beheerder voordat ze publiek verschijnen.

3 jury checks · meest recent 5 uur geleden
15 May 2026 5 jurors · onbeslist, onbeslist, kan, kan, onbeslist onbeslist
12 May 2026 3 jurors · kan, kan niet, kan onbeslist
11 May 2026 3 jurors · kan, kan niet, kan onbeslist status gewijzigd

Elke rij is een afzonderlijke jurycontrole. Juryleden zijn AI-modellen (identiteiten bewust neutraal gehouden). Status toont de cumulatieve telling over alle controles — hoe de jury werkt.

Meer in Judgment

Hebben we er één gemist?

We review weekly.