NetTIME

NetTIME predicts cell-type-specific transcription factor (TF) binding sites at base-pair resolution by leveraging next-generation sequencing data and enabling knowledge transfer from data-rich TFs and cell types to data-limited ones.

Key Features:

• Multitask Learning Framework: NetTIME employs a multitask learning strategy to share information across multiple TFs and cell types and improve prediction accuracy.
• High-Dimensional Embedding Vectors: The model trains high-dimensional embedding vectors that encode TF and cell-type identities to enable accurate transfer predictions within and beyond training datasets.
• Linear-Chain Conditional Random Field (CRF): NetTIME incorporates a linear-chain conditional random field (CRF) to classify base-pair-resolution binding predictions and avoid manual probability-thresholding.
• Base-Pair Resolution Predictions: Produces binding-site predictions at single-nucleotide (base-pair) resolution.
• Next-Generation Sequencing and Evaluation: Leverages next-generation sequencing data and uses standardized model evaluation criteria.
• Benchmark Performance: Demonstrates better performance compared to state-of-the-art methods Catchitt and Leopard under supervised and transfer learning conditions.

Scientific Applications:

• TF Binding Site Mapping: Provides precise binding site information across various cell types and conditions to support studies of gene regulation mechanisms.
• Transfer Learning for Less-Characterized TFs/Cell Types: Enables transfer predictions to study TFs and cell types with limited experimental data.

Methodology:

NetTIME trains high-dimensional TF and cell-type embedding vectors within a multitask learning framework and applies a linear-chain conditional random field (CRF) to classify base-pair-resolution binding predictions using next-generation sequencing data and standardized model evaluation criteria in both supervised and transfer learning settings.