Research map / Label-efficient and robust learning / Recognition and long-tail
Enhancing Transcription Factor Prediction through Multi-Task Learning (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a multi-task CNN framework on ESM-2 protein embeddings that jointly predicts transcription factors and TF-related domain classes, outperforming single-task and DeepTFactor/TFpredict baselines.
Problem
Accurate prediction of transcription factors from protein sequences, where existing tools (TFpredict, DeepTFactor) have limited accuracy or need large datasets.
Approach
Jointly train a shared 1-D CNN over ESM-2 sequence embeddings (with window-averaged condensation) for TF vs non-TF prediction and TF-related domain classification, exploiting shared information between the two tasks.
Key results
- Multi-task model reaches F1 0.9422, specificity 0.9253, sensitivity 0.9573, accuracy 0.9432 on TF prediction under five-fold cross-validation, versus DeepTFactor accuracy 0.9169.
- Multi-task learning improves TF-related domain classification accuracy to 0.9085 versus 0.8842 for single-task.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use multi-task learning instead of single-task learning | ESM-2 TF prediction framework · TF-related domain classification | F1 score: 0.8967 -> 0.9158 | Helped |
| use multi-task learning instead of single-task learning | ESM-2 TF prediction framework · TF-related domain classification | Accuracy: 0.8842 -> 0.9085 | Helped |
| use multi-task learning instead of single-task learning | ESM-2 TF prediction framework · TF prediction | F1 score: 0.9225 -> 0.9422 | Helped |
| use multi-task learning instead of single-task learning | ESM-2 TF prediction framework · TF prediction | Specificity: 0.8717 -> 0.9253 | Helped |
| use multi-task learning instead of single-task learning | ESM-2 TF prediction framework · TF prediction | Sensitivity: 0.9466 -> 0.9573 | Helped |
| use multi-task learning instead of single-task learning | ESM-2 TF prediction framework · TF prediction | Accuracy: 0.9252 -> 0.9432 | Helped |
| use multi-task learning instead of DeepTFactor | TF prediction · TF prediction | F1 score: 0.9171 -> 0.9422 | Helped |
| use multi-task learning instead of DeepTFactor | TF prediction · TF prediction | Specificity: 0.9056 -> 0.9253 | Helped |
| use multi-task learning instead of DeepTFactor | TF prediction · TF prediction | Sensitivity: 0.9283 -> 0.9573 | Helped |
| use multi-task learning instead of DeepTFactor | TF prediction · TF prediction | Accuracy: 0.9169 -> 0.9432 | Helped |
| add multi-task learning rather than single-task learning (ablation) | ESM-2 TF prediction framework · TF-related domain classification | F1 score: 0.8967 -> 0.9158 | Helped |
| add multi-task learning rather than single-task learning (ablation) | ESM-2 TF prediction framework · TF-related domain classification | Accuracy: 0.8842 -> 0.9085 | Helped |
Limitations
- Small student abstract; limited details on hyperparameter tuning and data splits.
- Comparison against only two existing TF predictors.
Experimental setup
Datasets: UniProt TF annotations UniProt TF-related domain annotations
Benchmarks: UniProt-derived TF prediction dataset
Metrics: F1 score specificity sensitivity accuracy
Baselines: TFpredict DeepTFactor Single-task learning
Builds on: ESM-2 DeepTFactor
Closest papers by meaning
- C3P: Contrastive promoter-protein pretraining yields representations capturing bacterial gene regulation (NeurIPS 2026 · similarity 0.83)
- Multi-modal Transfer Learning between Biological Foundation Models (NeurIPS 2024 · similarity 0.83)
- Context-Aware Regularization with Markovian Integration for Attention-Based Nucleotide Analysis (NeurIPS 2025 · similarity 0.83)
- What DNA Foundation Models Learn Beyond Sequence Composition (NeurIPS 2026 · similarity 0.83)
- Learning to Discover Regulatory Elements for Gene Expression Prediction (ICLR 2025 · similarity 0.83)
- Regulatory DNA Sequence Design with Reinforcement Learning (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.