Preprint
Machine Learning

CRISPR-DIPOFF: an interpretable deep learning approach for CRISPR Cas-9 off-target prediction

Md Toufikuzzaman(Department of Computer Science and Engineering, Bangladesh University of Engineering and Technology , Dhaka, 1205, Bangladesh), Md Abul Hassan Samee(Department of Integrative Physiology, Baylor College of Medicine , Houston, TX 77030, USA), M Sohel Rahman(Department of Computer Science and Engineering, Bangladesh University of Engineering and Technology , Dhaka, 1205, Bangladesh)
January 22, 2024Briefings in Bioinformatics
DOI

0

Citations

0

Influential Citations

Briefings in Bioinformatics

Venue

2024

Year

Abstract

Abstract CRISPR Cas-9 is a groundbreaking genome-editing tool that harnesses bacterial defense systems to alter DNA sequences accurately. This innovative technology holds vast promise in multiple domains like biotechnology, agriculture and medicine. However, such power does not come without its own peril, and one such issue is the potential for unintended modifications (Off-Target), which highlights the need for accurate prediction and mitigation strategies. Though previous studies have demonstrated improvement in Off-Target prediction capability with the application of deep learning, they often struggle with the precision-recall trade-off, limiting their effectiveness and do not provide proper interpretation of the complex decision-making process of their models. To address these limitations, we have thoroughly explored deep learning networks, particularly the recurrent neural network based models, leveraging their established success in handling sequence data. Furthermore, we have employed genetic algorithm for hyperparameter tuning to optimize these models’ performance. The results from our experiments demonstrate significant performance improvement compared with the current state-of-the-art in Off-Target prediction, highlighting the efficacy of our approach. Furthermore, leveraging the power of the integrated gradient method, we make an effort to interpret our models resulting in a detailed analysis and understanding of the underlying factors that contribute to Off-Target predictions, in particular the presence of two sub-regions in the seed region of single guide RNA which extends the established biological hypothesis of Off-Target effects. To the best of our knowledge, our model can be considered as the first model combining high efficacy, interpretability and a desirable balance between precision and recall.

Analysis

Why This Paper Matters

CRISPR-Cas9 is a revolutionary genome-editing tool with immense potential, but off-target effects pose significant safety risks. Accurate prediction of off-target sites is critical for therapeutic applications and basic research. While deep learning has improved prediction accuracy, previous models often suffer from a precision-recall trade-off and lack interpretability, making them less reliable and harder to trust in clinical settings.

CRISPR-DIPOFF addresses these gaps by not only improving performance but also providing interpretable insights. The identification of two sub-regions in the seed region of sgRNA extends the biological understanding of off-target mechanisms, which could inform better guide design and reduce unintended edits. This dual focus on efficacy and interpretability sets a new standard for computational tools in genomics.

Technical Contributions

  • RNN-based architecture: Leverages recurrent neural networks, which are well-suited for sequence data, to model the DNA and RNA sequences involved in off-target binding.
  • Genetic algorithm hyperparameter tuning: Automates the search for optimal hyperparameters, leading to better model performance than manual tuning or grid search.
  • Integrated gradients for interpretability: Applies a gradient-based attribution method to identify which parts of the input sequence contribute most to predictions, providing a window into the model's reasoning.
  • Precision-recall balance: Explicitly addresses the trade-off by optimizing for both metrics, resulting in a more clinically useful model.
  • Biological insight: The interpretation reveals two sub-regions within the seed region, refining the existing hypothesis about off-target effects.

Results

The paper reports significant performance improvements over current state-of-the-art methods, though specific metrics (e.g., AUC, precision, recall) are not detailed in the abstract. The key achievement is a better balance between precision and recall, meaning fewer false positives and false negatives. The interpretability analysis provides a detailed understanding of the factors driving predictions, with the discovery of two sub-regions in the seed region being a notable finding.

Significance

CRISPR-DIPOFF represents a step forward in making deep learning models for genomics both powerful and interpretable. By combining high efficacy with interpretability, it addresses a common criticism of deep learning as a 'black box.' This work could influence future model development in other biological sequence prediction tasks, encouraging the integration of interpretability from the outset. The biological insights may also guide experimental validation and improve CRISPR design tools, ultimately enhancing the safety and efficacy of genome editing.

deep-learningcrisproff-target-predictionrecurrent-neural-networksgenetic-algorithminterpretabilityintegrated-gradientsbioinformatics
Neural Networks

Related Papers