Abstract
BACKGROUND: Transcription factor (TF)-dependent DNA demethylation is associated with generation of specific DNA methylation profiles in normal cellular development and disease, although only a small fraction of TFs are known to promote DNA demethylation. RESULTS: Here, we systematically predicted which TFs have DNA-demethylation-promoting activity. Experiments with deletion mutants of the TFs RUNX1 and SPI1 revealed that this activity is associated with a relatively long intrinsically disordered region (IDR). Examination of the IDRs from eight TFs previously confirmed to have such activity revealed that at least one IDR was active in each TF. We constructed a Random Forest classifier based on 25 numeric physicochemical features extracted from length-controlled 26 positive (active) and 32 negative (inactive) IDRs. Four key features- aromaticity, aliphatic index, fractional charge ratio, and side chain hydrophobic density-were identified as the most informative contributors to prediction of positive IDRs. A model based on these features achieved an area under the receiver operating characteristic curve of 0.84, with an optimized decision threshold of 0.303. Applying this model to all TFs, we predicted 959 of 2364 IDRs to be positive, corresponding to 825 of 1308 TFs. The model correctly identified all of 14 previously validated positive TFs. The predicted positive TFs showed significant enrichment of Gene Ontology terms related to morphogenesis and development and may be clinically relevant to certain cancer types. CONCLUSION: The developed model with high predictive performance and the predicted TFs with DNA-demethylation-promoting activity will be useful for further analysis of TFs involved in generation of DNA methylation profiles in normal cell development and disease.