AI cannot cure cancers, Alzheimer.

Home page
AI is useless.

New AI tools = ProRNA3D and Alphafold-3 gave only wrong proteins or RNA structures.  ← AI drug discovery is hopeless

(Fig.1)   AI's predicted proteins or RNAs always contain more than 80% wrong structures.

New AI tools or Alphafold-3 cannot predict protein-RNA structures.

The 9th, 11th, last paragraphs of this hyped news (9/16/2025) say

"ProRNA3D-single (= this new AI tool ) can generate 3D structural models of viral RNA interacting with proteins"  ← fake news

"Even recent breakthrough AI models developed by Google DeepMind and others have fallen far short of accurately predicting and modeling protein-RNA complex 3D structures,"  ← AI, Alphafold are useless.

"We should constantly remind ourselves the problem is far from being solved"  ← This new AI tool is also useless for drug discovery ( this-3rd-paragraph ).

More than 80% of protein structures predicted by AI were false.  ← hopeless AI

Today's AI trained on wrong protein-RNA datasets in PDB gave only wrong structures (= fnat score < 0.2   ← 1 is right  = 80% error rates ).

↑ This research paper ↓

p.8-2nd-paragraph says "To train our new method, ProRNA3D-single, we collect X-ray crystal structures of protein-RNA complexes having resolution up to 3.5 Å (= non-atomic bad resolution or wrong structures were used as their training datasets ) that are deposited to the PDB"

p.8-3rd-paragraph says "Using protein-RNA complexes published more recently than any training- and validation-set targets.. resulting in a total of 39 protein-RNA complexes for the test set,"  ← Recently-obtained experimental protein-RNA structures were used for the test.

p.9-2nd-paragraph says "The fnat scores lie in range of [ 0-1 ], and the higher score represents better accuracy of the predicted complex structural model (= fnat score closer to 1 is more accurate,  this-p.6-2nd-last-paragraph )"

p.10-1st-paragraph says "ProRNA3D-single attains the best fnat scores compared to the state-of-the-art methods, attaining 41.03% successful prediction having fnat > 0.2, which is 25.64%, 25.64%, and 17.95% higher than RF2NA, RF2AA, and AF3 (= Alphafold-3 ), respectively, where a prediction is considered "successful" if the predicted structural model has fnat>0.2"

More than 80% of AI prediction was wrong (= fnat = 0.2 )

The present bad datasets keep AI hopeless for drug discovery forever, spawning overhyped fake news.

↑ This new AI tool called ProRNA3D gave only wrong protein-RNA structures whose more than 80% structures (= fnat score 0.2 between 0 and 1  = 1 is accurate ) were wrong.

↑ Alphafold-3 (= AF3 ) gave worse results where about 80% of its predicted structures contained more than 80% wrong protein or RNA structures.

Alphafold-3 (= AF-3 ) also cannot give right structures (= fnat score < 1 ) even of proteins ( this-p.1 ).

As a result, today's AI tools cannot predict accurate structures of proteins nor RNAs, so drug discovery is impossible due to today's bad experimental datasets, just spawning overhyped fake news.

Feel free to link to this site.