MIT's CheMeleon model predicts molecular behavior from limited data

The newly developed open-source CheMeleon, pre-trained on data from one million molecules, accelerates drug and material discovery even with a small amount of experimental data. The model demonstrates superiority over other methods in 58 test cases, reducing experimental costs.