Deep Learning Basecalling for Fluorescent DNA Sequencing Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Next-generation sequencing techniques generate massive amounts of data with challenging signal processing issues, including crosstalk and loss of synchrony in cluster molecules, leading to inefficiencies and inaccuracies in basecalling, which are not adequately addressed by existing machine learning models.
Innovation Solution
Employing a combination of image preprocessing algorithms, improved cluster detection using convolutional neural networks (CNN), and attention-based or 1D convolution-based neural networks for parallel processing, along with transformer neural networks to enhance cluster detection and basecalling accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional algorithms are used for basecalling, then computational resources and time are consumed, but basecalling accuracy remains insufficient
Solution Approach 1:
The patent transforms the basecalling problem from traditional sequential processing to parallel processing by changing the computational parameters. The 1D convolutional neural network processes multiple clusters simultaneously, and the attention mechanism allows parallel computation of attention scores for different positions, fundamentally changing how the basecalling task is executed and achieving both high accuracy and speed.
Solution Approach 2:
The patent replaces traditional mechanical sequential processing algorithms with neural network-based parallel processing systems. The convolutional neural network layers and attention mechanisms substitute for conventional step-by-step basecalling algorithms, enabling simultaneous processing of multiple data points and achieving superior performance in both accuracy and speed.
2Measurement precision
If existing machine learning models (CNN+RNN) are used, then basecalling accuracy is improved, but computational complexity and resource requirements increase
Solution Approach 1:
The patent segments the basecalling task into distinct computational stages: feature extraction through 1D convolutional layers, attention score computation, and final base calling. This segmentation allows each component to be optimized independently and processed in parallel, reducing overall computational complexity while maintaining high accuracy.
Solution Approach 2:
The patent introduces a new dimensional approach by using 1D convolutions over the sequence dimension and attention mechanisms that operate across positional dimensions simultaneously. This dimensional transformation enables parallel processing of multiple clusters and positions, reducing computational complexity compared to traditional 2D CNN+RNN architectures.
3Productivity
If parallel processing is implemented, then basecalling speed is improved, but handling of crosstalk and signal synchronization challenges becomes more difficult
Solution Approach 1:
The patent introduces an attention mechanism as an intermediary between the extracted features and the final base calling decision. This attention layer computes attention scores that automatically weigh and adjust for crosstalk and synchronization issues in the parallel processing signals, mediating the complex interactions between multiple clusters and enabling accurate parallel basecalling despite signal challenges.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods and systems for determining a plurality of sequences of nucleic acid (e.g., DNA) molecules in a sequencing-by-synthesis process are provided. In one embodiment, the method comprises obtaining images of fluorescent signals obtained in a plurality of synthesis cycles. The images of fluorescent signals are associated with a plurality of different fluorescence channels. The method further comprises preprocessing the images of fluorescent signals to obtain processed images. Based on a set of the processed images, the method further comprises detecting center positions of clusters of the fluorescent signals using a trained convolutional neural network (CNN) and extracting, based on the center positions of the clusters of fluorescent signals, features from the set of the processed images to generate feature embedding vectors. The method further comprises determining, in parallel, the plurality of sequences of DNA molecules using the extracted features based on a trained attention-based neural network.