Boosting Single-Cell RNA Sequencing Analysis with Simple Neural Attention
Oscar A Davalos1, A Ali Heydari2,3, Elana J Fertig4
1Quantitative and Systems Biology Graduate Program, University of California, Merced, CA, USA.
Biorxiv : the Preprint Server for Biology
|July 3, 2023
Summary
scANNA is a new interpretable deep learning model for single-cell RNA sequencing (scRNAseq) analysis. It uses gene importance learned from neural attention for downstream tasks, improving efficiency and results.
Area of Science:
- Computational Biology
- Genomics
- Machine Learning
Background:
- Current deep learning (DL) models for single-cell RNA sequencing (scRNAseq) analysis lack interpretability and require task-specific training.
- Existing scRNAseq analysis pipelines are often disjointed, necessitating separate models for different analytical stages.
Approach:
- We introduce scANNA, an interpretable DL model for scRNAseq data that utilizes neural attention mechanisms.
- scANNA learns gene associations and importance during training, enabling direct application to downstream analyses without retraining.
Key Points:
- The interpretability of scANNA allows for the identification of gene importance, facilitating tasks like marker selection and cell-type classification.
- scANNA achieves performance comparable to or exceeding state-of-the-art methods on standard scRNAseq tasks without task-specific training.
- The model enhances scRNAseq analysis by reducing the need for extensive prior knowledge and eliminating the requirement for multiple specialized models.
Conclusions:
- scANNA offers a unified and interpretable framework for scRNAseq analysis, streamlining research workflows.
- This approach empowers researchers to gain meaningful insights efficiently, saving time and computational resources.
