13: Machine Learning & Deep Learning in Modern Genomics

Chapter 13: Machine Learning & Deep Learning in Modern Genomics

From predicting variant pathogenicity and alternative splicing to forecasting gene expression directly from unannotated DNA sequences, Machine Learning (ML) and Deep Learning (DL) have become the dominant methodological drivers of contemporary bioinformatics.

1. 1D CNNs as Automated Motif Detectors

A 1D convolutional filter sliding across a one-hot encoded DNA sequence acts as a mathematical equivalent to a Position Weight Matrix (PWM). The weights of the first convolutional layer directly learn position-specific nucleotide affinities, discovering regulatory binding motifs without prior manual feature engineering.

2. Genomic Foundation Models

Modern architectures adapt large language models to raw genomics, treating genomes as biological text. Models like DNABERT, Nucleotide Transformer, and HyenaDNA are pre-trained via masked language modeling across thousands of vertebrate genomes, learning complex long-range promoter-enhancer interactions spanning hundreds of kilobases.

3. PyTorch Architecture for Sequence Classification

import torch
import torch.nn as nn

class GenomicMotifCNN(nn.Module):
    def __init__(self, num_filters=32, kernel_size=15):
        super(GenomicMotifCNN, self).__init__()
        # Input shape: (Batch, 4 channels [A, C, G, T], Sequence_Length)
        self.conv1 = nn.Conv1d(in_channels=4, out_channels=num_filters, kernel_size=kernel_size)
        self.relu = nn.ReLU()
        self.pool = nn.MaxPool1d(kernel_size=4)
        self.flatten = nn.Flatten()
        self.fc = nn.Linear(num_filters * 21, 1) # Assumes 100 bp input -> ~21 post-pooling
        self.sigmoid = nn.Sigmoid()
        
    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.pool(x)
        x = self.flatten(x)
        out = self.sigmoid(self.fc(x))
        return out

model = GenomicMotifCNN()
print("Initialized Genomic CNN:\n", model)