Decoding Circadian Gene Regulation Using a Novel Machine Learning Framework

Overview

Circadian clocks are internal 24 -hour rhythms that synchronize body functions with day -night cycles. At the molecular level, they consist of gene networks that control gene oscillations, governing nearly all physiological processes. Circadian rhythm disruption is increasingly linked to disease, particularly cancer, yet circadian patterns in different tissues, their regulators, and the mechanisms and effects of circadian gene disruptions remain largely unknown. To address this, I studied genes’ circadian programs across healthy tissues and modeled how disruptions in clock regulators alter these rhythms. Using RNA -seq data from 64 baboon tissue -types, I identified 13,211 significantly oscillating genes (false discovery rate < 0.2) and quantified their oscillat ion patterns, including amplitude and peak-transcription times, revealing coordinated "wake-up" times across tissues. By integrating public RNA -seq and ChIP-seq data from ENCODE, I developed a Random Forest machine learning framework, CYFOR, to infer regul atory networks governing genes’ circadian patterns, identifying 37 new circadian regulators beyond the 9 known core clock genes. Follow-up experimental studies on a CYFOR-predicted circadian regulator, MYC – a key oncogene driving 40% of cancers – revealed that it binds to similar DNA sequences as core clock gene ARNTL (p-value < 0.0001). Additionally, RNA-seq analyses in a human neuroblastoma cell line, SHEP, show that MYC overexpression disrupted rhythms of 88.5% genes oscillating in normal conditions. Th ese data validate my computational prediction and suggest a cancer - causing mechanism where MYC disrupts normal circadian rhythms by competing with core clock genes on their DNA binding sites, demonstrating CYFOR’s ability to model circadian disruption in human disease.