Apple researchers develop SimpleFold, a lightweight AI for protein folding prediction

Apple researchers have developed SimpleFold, a lightweight AI model for protein structure prediction that offers a more efficient alternative to Google DeepMind's AlphaFold. While AlphaFold reduced prediction time from months to minutes, it requires expensive calculations and strict frameworks. SimpleFold uses flow matching models instead of computationally heavy designs like multiple sequence alignments and triangle updates. Testing across model sizes (100M to 3B parameters) showed SimpleFold achieves competitive performance with leading models while requiring less computational power. The model achieved over 95% of RoseTTAFold2/AlphaFold2's performance, demonstrating that effective protein folding can be accomplished with general-purpose architecture.