Transcription factors (TFs) bind DNA in a sequence-specific manner to regulate gene expression. To understand gene regulation and regulatory networks, it is essential not only to map differences in TF-DNA binding in the genome but also to determine the biophysical basis of specificity. The former is possible using comprehensive in vivo assays, and the latter requires focused thermodynamic analyses. However, it has been challenging to relate in vivo binding data to in vitro-derived thermodynamic affinities. My work will show that neural networks can extract thermodynamic affinities de novo from genomic occupancy profiles, enabling massive in silico experiments to decipher sequence influences on intrinsic affinity and in vivo occupancy.