Why AI still cannot parse the complexity of the human genome
AI summary
Only about 2% of the human genome consists of protein-coding genes; what is more complex is how genes are regulated, and genomic foundation models such as Evo 2, Genos, and AlphaGenome merely learn correlations from known cases and cannot provide genuine mechanistic understanding. The article points out that the genome is not a blueprint or an algorithm, and its regulatory logic (for example, transcription factors working in pairs or groups, and acting in opposite ways in different cell types) is difficult to fit into the input-output framework assumed by AI models.
Body
The English body is being prepared. You can open the original source.