Complex-valued deep learning for genomic sequencing

Abstract

A pathogen such as SARS-CoV-2 is spreading fast due to the lack of accuracy in the detection tools that are currently used in practice, this is likely to hold true for all future pathogens. Future integration of virus genomic data may enable forecasting of the spatial spread and ignition/decline of epidemics. For example, nucleotide sequences such“as GTTATGGGACCAATTG” and amino acid sequences such as“RGFGDSVEEALSEAREHLKNGTCGLVE” can be represented in frequency domain much like electromagnetic signals. Recently, the applications of complex-valued deep learnings have become wider in the field of MRI signal processing speech enhancement, wind prediction, image classification and segmentation, then why they could not be applied to genomics? The aforementioned deep-learning here is mainly about a combination of Convolution Neural Network (CNN) and Multilayer Perceptron (MLP)for both real-valued as a baseline and complex-valued architecture. Complex-valued deep learnings are the most compatible model for wave related in the wave signal, phase components represent the time course or position difference, whereas amplitude denotes the energy or power of the wave. Hence, wave information can be fully extracted when training it with complex-valued deep learnings rather than real-valued deep learnings. If we keep the same number of neurons (for fully connected layers) or kernels (for convolutional layers) for both real-valued deep learning and complex-valued deep learning , it will result in the complex-valued deep learning having higher capacity or used memory in the computer’s RAM (Random Access Memory) than their opposed real-valued deep learning, especially Real valued Multilayer Perceptron (RVMLP), which is unfair as we can consider that the complex plane is isomorphic or have a similar shape to 2D real vectors. Which means one complex-valued parameter is equivalent to two real-valued parameters and the real-valued parameters should be roughly the same for both sides to provide fair comparisons.