AI Base Caller with Contextual Awareness for Sequencing Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks face memory limitations when processing large images, leading to loss of local details and global contextual information due to downsampling or patch-based processing, which results in inaccurate base calling.
Innovation Solution
An artificial intelligence-based base caller with contextual awareness is developed, utilizing an intensity contextualization unit that generates intensity context data from sequencing images, which is then incorporated into the neural network's processing, allowing it to consider both local patch details and global contextual information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If convolutional neural networks process large images, then base calling accuracy improves, but memory requirements increase exponentially
Solution Approach 1:
The patent divides the large sequencing image into multiple smaller patches that are processed independently through the convolutional neural network. Each patch is processed separately to generate base calls, which are then assembled to form the complete sequence. This segmentation approach reduces memory requirements by avoiding the need to store and process the entire large image at once, while maintaining base calling accuracy through comprehensive patch coverage.
Solution Approach 2:
The patent introduces a contextual awareness dimension by generating contextual information from the entire sequencing image and integrating it with patch-level predictions. This adds a global context dimension to the otherwise local patch-based processing, allowing the system to maintain accuracy without requiring the neural network to process the complete large image simultaneously in its standard representation.
2Quantity of substance
If downsampling is applied to reduce memory usage, then memory requirements decrease, but local details and global contextual information are lost
Solution Approach 1:
The patent segments the sequencing image into multiple patches that retain fine local details, while separately extracting global contextual information from the complete image. This segmentation strategy allows the system to process local regions at high resolution without requiring the entire large image to be stored in memory, thereby preserving both local details and global context.
Solution Approach 2:
The patent introduces contextual information as an intermediary element that bridges local patch processing and global image understanding. This contextual representation serves as a mediator that provides global awareness to the patch-level predictions without requiring the neural network to directly process the complete large image, thus preventing information loss while reducing memory requirements.
3Quantity of substance
If patch-based processing is used, then memory requirements decrease, but contextual information from outside the patch is lost
Solution Approach 1:
The patent introduces contextual information as an intermediary that externalizes the contextual awareness from the patch-based processing. This contextual representation acts as a mediator that provides global contextual information to the patch-level predictions, allowing the system to maintain memory efficiency while recovering the contextual information that would otherwise be lost in isolated patch processing.
Solution Approach 2:
The patent replaces the mechanical approach of processing the entire large image through the neural network with a hybrid system that combines patch-based neural network processing with a separate contextual information generation mechanism. This substitution allows the system to achieve both memory efficiency and contextual awareness without requiring the neural network to directly process the complete large image.
Data Source
AI summary
A neural network processes sequencing images on a patch-by-patch basis for base calling. The sequencing images depict intensity emissions of a set of analytes. The patches depict the intensity emissions for a subset of the analytes and have undiverse intensity patterns due to limited base diversity. The neural network has convolution filters that have receptive fields confined to the patches. The convolution filters detect intensity patterns in the patches with losses in detection due to the undiverse intensity patterns and confined receptive fields. An intensity contextualization unit determines intensity context data based on intensity values in the images. The data flow logic appends the intensity context data to the sequencing images to generate intensity contextualized images. The neural network applies the convolution filters on the intensity contextualized images and generates base call classifications. The intensity context data in the intensity contextualized images compensates for the losses in detection.