#37 · Epigenomics

Explicit reference criteria for evaluating biological model outputs and representations

The dominant thread is whether model outputs or representations can be assessed against explicit biological or measurement references rather than accepted as opaque. HiFold-Mamba, a Stage 1 Registered Report, proposes a reverse-complement-equivariant state-space model that predicts 3D genome contact maps from DNA and tests whether an insulation-based map-level constraint improves sequence-based prediction. The cryo-EM preprint introduces ramped phase randomization, a low-pass filter variant used for map comparison and neural network data augmentation, with an open-source Python implementation. The explicit framing of evaluation criteria is what distinguishes these two works. The EEG study evolves spiking networks with calcium-dynamic neurons and synaptic delays for seizure detection on CHB-MIT and Siena data, and is less directly aligned with this evaluative thread. All three are preprints, so the evidence remains preliminary.