[Paper Review] Deep learning in bioinformatics: introduction, application, and perspective in big data era

Yu Li, Chao Huang+4 more|arXiv (Cornell University)|Feb 28, 2019

Machine Learning in Bioinformatics198 references4 citations

TL;DR

This review introduces deep learning in bioinformatics, offering an accessible overview of key architectures—such as CNNs, RNNs, GNNs, GANs, and VAEs—and provides practical, TensorFlow/Keras-based implementations across five bioinformatics domains and four data types. It addresses challenges like overfitting and interpretability, with code publicly available, positioning deep learning as essential for big data-driven biological research.

ABSTRACT

Deep learning, which is especially formidable in handling big data, has achieved great success in various fields, including bioinformatics. With the advances of the big data era in biology, it is foreseeable that deep learning will become increasingly important in the field and will be incorporated in vast majorities of analysis pipelines. In this review, we provide both the exoteric introduction of deep learning, and concrete examples and implementations of its representative applications in bioinformatics. We start from the recent achievements of deep learning in the bioinformatics field, pointing out the problems which are suitable to use deep learning. After that, we introduce deep learning in an easy-to-understand fashion, from shallow neural networks to legendary convolutional neural networks, legendary recurrent neural networks, graph neural networks, generative adversarial networks, variational autoencoder, and the most recent state-of-the-art architectures. After that, we provide eight examples, covering five bioinformatics research directions and all the four kinds of data type, with the implementation written in Tensorflow and Keras. Finally, we discuss the common issues, such as overfitting and interpretability, that users will encounter when adopting deep learning methods and provide corresponding suggestions. The implementations are freely available at \url{https://github.com/lykaust15/Deep_learning_examples}.

Motivation & Objective

• To provide a comprehensive, accessible introduction to deep learning for bioinformatics researchers unfamiliar with neural networks.
• To demonstrate practical deep learning applications across diverse bioinformatics domains, including sequence analysis, structure prediction, and biomolecular function prediction.
• To address common challenges in adopting deep learning, such as overfitting, interpretability, and computational demands, with actionable solutions.
• To offer open-source, reproducible implementations in TensorFlow and Keras for eight representative use cases to accelerate method adoption.
• To position deep learning as a transformative tool for handling the growing complexity and scale of biological big data.

Proposed method

• Introduces deep learning fundamentals starting from shallow neural networks, progressing through CNNs, RNNs, and advanced architectures like GANs and VAEs.
• Employs a tiered educational approach: from basic concepts to state-of-the-art models, ensuring accessibility for non-experts.
• Applies deep learning techniques to five research areas: sequence analysis, structure prediction, biomolecular property/function prediction, biomedical imaging, and systems biology.
• Uses four data types—genomic sequences, structural data, omics profiles, and medical images—across the provided examples.
• Implements models using TensorFlow and Keras, with code hosted on GitHub for reproducibility and community reuse.
• Proposes model compression techniques—pruning, knowledge distillation, low-rank factorization, and compact filters—to reduce computational cost and memory usage.

Experimental results

Research questions

• RQ1 Which deep learning architectures are most effective for solving key problems in bioinformatics, such as sequence variant prediction and protein structure modeling?
• RQ2 How can deep learning models be made more interpretable and robust to overfitting in small or noisy biological datasets?
• RQ3 What strategies can mitigate catastrophic forgetting when updating models with new biological data without retraining from scratch?
• RQ4 How can model compression techniques improve deployment of deep learning in resource-limited biomedical computing environments?
• RQ5 What are the practical, implementable workflows for integrating deep learning into existing bioinformatics pipelines?

Key findings

• Deep learning models, particularly CNNs and RNNs, achieve state-of-the-art performance in tasks like protein secondary structure prediction, transcription factor binding site detection, and DNA variant effect prediction.
• Generative models such as VAEs and GANs show promise in modeling complex biological sequences and generating synthetic but biologically plausible data.
• Graph neural networks (GNNs) and message-passing architectures effectively model biomolecular interactions and systems-level biological networks.
• Model compression techniques like pruning and knowledge distillation significantly reduce model size and inference time, enabling deployment on edge devices and in clinical settings.
• The integration of deep learning with Bayesian inference (e.g., in DLBI) enhances super-resolution microscopy, accelerating image reconstruction without sacrificing accuracy.
• The authors demonstrate that transfer learning and continual learning strategies can reduce catastrophic forgetting, preserving performance on prior data while adapting to new biological knowledge.