Applications of Deep Learning and Reinforcement Learning to Biological Data (1711.03985v2)

Abstract: Rapid advances of hardware-based technologies during the past decades have opened up new possibilities for Life scientists to gather multimodal data in various application domains (e.g., Omics, Bioimaging, Medical Imaging, and [Brain/Body]-Machine Interfaces), thus generating novel opportunities for development of dedicated data intensive machine learning techniques. Overall, recent research in Deep learning (DL), Reinforcement learning (RL), and their combination (Deep RL) promise to revolutionize Artificial Intelligence. The growth in computational power accompanied by faster and increased data storage and declining computing costs have already allowed scientists in various fields to apply these techniques on datasets that were previously intractable for their size and complexity. This review article provides a comprehensive survey on the application of DL, RL, and Deep RL techniques in mining Biological data. In addition, we compare performances of DL techniques when applied to different datasets across various application domains. Finally, we outline open issues in this challenging research area and discuss future development perspectives.

Applications of Deep Learning and Reinforcement Learning to Biological Data

The paper "Applications of Deep Learning and Reinforcement Learning to Biological Data" provides an exhaustive and critical review of how modern AI techniques, specifically Deep Learning (DL) and Reinforcement Learning (RL), contribute to the interpretation and analysis of complex biological datasets. The work explores the multifaceted domains of Omics, Bioimaging, Medical Imaging, and Brain/Body-Machine Interfaces, exploring the potential and existing implementations of these computational methodologies.

Overview of Techniques

The paper categorizes DL into several architectures, including Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Deep Autoencoders, and Boltzmann Machines, each addressing unique challenges posed by biological data. RL is discussed through its fundamental tenets of learning via interaction, with DL enhancing its applicability to vast and dynamic data environments.

Applications across Biological Domains

Omics

DL and RL methodologies are particularly transformative in the Omics domain, which deals with high-dimensional data such as genomics, proteomics, and metabolomics. The paper details applications such as gene expression profiling, DNA methylation state prediction, and protein structure determination. The employment of CNNs and DNNs in these tasks showcases superior performance compared to traditional machine learning techniques in both classification and predictive analytics, evidenced by higher accuracy in tasks like splice junction prediction and compound-protein interaction.

Bioimaging

In the field of bioimaging, the paper highlights the efficacy of DL in tasks such as cell classification and segmentation, leveraging CNNs to decode complex patterns in microscopy images. The capability of DL to perform pixel-wise analyses bolsters its utility in identifying mitosis in histology images and nuclei detection.

Medical Imaging

Medical imaging sees a robust deployment of DL and RL, where tasks such as segmentation, denoising, and anomaly detection are paramount. Techniques like CNNs have been instrumental in tasks ranging from tumor detection in brain MRIs to segmenting glandular structures in colon histology images. The inclusion of RL frameworks further facilitates adaptive learning and decision-making processes in image interpretation.

Brain/Body-Machine Interfaces

The application of DL and RL in Brain/Body-Machine Interfaces (BMI) highlights their role in decoding motor imagery from EEG signals and detecting anomalies such as seizures. Architectural models like multi-channel CNNs and DBNs have demonstrated substantial improvements in classification accuracy, proving beneficial for real-time applications in neuroprosthetics.

Performance Analysis

The paper provides a meticulous performance comparison across DL techniques applied to tasks in each biological domain. This comparison illustrates the enhanced predictive power and accuracy of DL architectures over conventional approaches, underscoring their adaptability and precision in handling complex data types inherent to biological applications.

Open Issues and Future Perspectives

Despite their advantages, the paper identifies several challenges remaining in DL and RL applications, including computational demands and interpretability concerns, commonly termed the "black box" issue. Future research directions proposed include advancing theoretical underpinnings, improving computational efficiency, and fostering greater integration with practical biological investigations. The suggestion to explore distributed and cloud computing for large-scale data processing also stands crucial.

Conclusion

This review signifies the crucial intersection of machine learning with biological sciences, presenting DL and RL as indispensable tools in modern biological research. The detailed coverage of applications, challenges, and future directions positions the paper as a valuable resource for researchers aiming to harness AI's full potential in life sciences. The paper's contributions lay a comprehensive groundwork for ongoing innovation and application of these transformative technologies in diverse biological domains.