Machine Learning-Assisted Decoding of Temporal Transcriptional Dynamics via Fluorescent Timer

Investigating the temporal dynamics of gene expression is crucial for understanding gene regulation across various biological processes. The Fluorescent Timer protein (Timer) offers a valuable tool for such studies, exemplified by our Foxp3 Timer-of-cell- kinetics-and-activity (Tocky), which facilitates the analysis of Foxp3 dynamics at the single-cell level. However, the complexity of Timer fluorescence profiles has limited its application, necessitating novel analytic approaches. Here, we introduce an integrative method that combines molecular biology with machine learning (ML) to elucidate enhancer-specific regulation of transcriptional dynamics. Our approach is demonstrated by mutating the enhancer, Conserved Non-coding Sequence 2 (CNS2), of the Foxp3 Timer transgene in Foxp3-Tocky embryos via CRISPR. Technologically, we have developed machine learning tools that employ Random Forest and Convolutional Neural Networks with image conversion techniques, including Gradient-weighted Class Activation Mapping. These independent ML models effectively elucidated CNS2-specific regulation of Foxp3 transcription and underscored the roles of CNS2 in regulating the autoregulatory loop of Foxp3 transcription. In conclusion, our study reveals previously unrecognized roles of CNS2 in Foxp3 transcriptional dynamics, showcasing the potential of the CRISPR Tocky assay as an advanced method to understand transcriptional dynamics in vivo.