FENNEC: Fine-Tuned Ensemble Neural Networks Accelerate Chemically Modified siRNA Design and Screening - Dataset

Small interfering RNAs (siRNAs) are a clinically validated therapeutic modality, but designing potent chemically modified siRNAs remains costly and iterative due to limited public data. Computational prediction of siRNA efficacy can accelerate rational design and preclinical development. However, despite the importance of chemical modifications in therapeutic performance, current machine-learning methods either do not account for this or generalize poorly beyond their training data.Here, we present FENNEC (Fine-Tuned Ensemble of Neural Networks for siRNA Efficiency Characterization), a machine-learning framework for predicting chemically modified siRNA activity. We curate the largest publicly available patent-derived dataset of modified siRNAs from 42 patents using OCR-based table extraction and stringent quality filtering. FENNEC combines temporal convolutional networks with thermodynamic features, experimental covariates, and RNA foundation-model embeddings to capture sequence- and transcript-level determinants of efficacy. We show that foundation-model embeddings encode transferable biological information, particularly benefiting prediction in data-scarce settings. This contains the data and weights used for this project. Goes together with marsico-lab/fennec repository.