Evaluate the utility of advanced DNA sequence-derived features for predicting 3D chromatin interactions

Determine the predictive value of advanced DNA sequence feature sets, including the 147 sequence feature encodings provided by the iLearnPlus platform, for predicting three-dimensional chromatin interaction patterns such as topologically associating domain (TAD) boundaries and chromatin loops, in comparison to conventional sequence encodings (one-hot and k-mer) and other feature extraction approaches.

Background

The review highlights that most current predictors of 3D genome organization rely on conventional sequence encodings (e.g., one-hot and k-mer) or epigenomic annotations, while newer sequence-based feature engineering frameworks are emerging. iLearnPlus provides 147 feature sets and multiple machine learning and deep learning pipelines for sequence feature extraction and modeling, but their relevance to 3D genome prediction tasks has not been systematically assessed.

Establishing the informativeness of these advanced sequence-derived features for predicting TAD boundaries, loops, and other interaction signatures would clarify whether sequence-only models can match or complement epigenomic-driven approaches, potentially improving cross-cell-type generalization and interpretability.