This paper presents an innovative approach to predicting the relative populations of protein conformations using AlphaFold 2 (AF2), an AI-powered method that has revolutionized protein structure prediction. While AF2 excels in predicting ground state conformations, it is limited in its ability to predict conformational landscapes. The authors demonstrate that by subsampling multiple sequence alignments (MSAs), AF2 can directly predict the relative populations of different protein conformations. They tested this method on two proteins, Abl kinase and granulocyte-macrophage colony-stimulating factor (GMCSF), with varying amounts of available sequence data, achieving over 80% accuracy in predicting changes in their relative state populations. The subsampling approach is particularly effective in qualitatively predicting the effects of mutations or evolution on conformational landscapes and well-populated states of proteins. This method offers a fast and cost-effective way to predict conformational distributions at single-point mutation resolution, making it valuable for pharmacology, experimental result analysis, and evolutionary studies. The results highlight the strong potential of AF2 for predicting changes in protein conformational ensembles, which will have significant impacts on biophysics and drug discovery.This paper presents an innovative approach to predicting the relative populations of protein conformations using AlphaFold 2 (AF2), an AI-powered method that has revolutionized protein structure prediction. While AF2 excels in predicting ground state conformations, it is limited in its ability to predict conformational landscapes. The authors demonstrate that by subsampling multiple sequence alignments (MSAs), AF2 can directly predict the relative populations of different protein conformations. They tested this method on two proteins, Abl kinase and granulocyte-macrophage colony-stimulating factor (GMCSF), with varying amounts of available sequence data, achieving over 80% accuracy in predicting changes in their relative state populations. The subsampling approach is particularly effective in qualitatively predicting the effects of mutations or evolution on conformational landscapes and well-populated states of proteins. This method offers a fast and cost-effective way to predict conformational distributions at single-point mutation resolution, making it valuable for pharmacology, experimental result analysis, and evolutionary studies. The results highlight the strong potential of AF2 for predicting changes in protein conformational ensembles, which will have significant impacts on biophysics and drug discovery.