DeeReCT-APA Deep Learning for Alternative Polyadenylation Site Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the usage of alternative polyadenylation sites (PAS) in genes can only handle two sites at a time, ignoring competition among multiple sites, which limits their ability to quantify the usage of all competing PAS simultaneously.
Innovation Solution
The DeeReCT-APA model uses a deep learning architecture with Bidirectional Long Short Term Memory (Bi-LSTM) layers to process genomic sequences centered around multiple PAS, allowing for simultaneous prediction of usage levels across all competing sites by modeling interactions between different PAS.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If current methods handle only two PAS at a time, then the prediction process is simple, but the ability to quantify usage of all competing PAS simultaneously is limited
Solution Approach 1:
The patent divides the prediction task into multiple independent binary classification problems, where each PAS is evaluated separately against all other PAS. This segmentation allows the system to handle multiple competing PAS simultaneously while maintaining manageable complexity through modular processing of each site's usage probability
Solution Approach 2:
The patent merges the predictions from multiple binary classification models into a unified set of usage probabilities for all PAS. By combining the outputs of individual PAS predictions and normalizing them to sum to 1, the system achieves simultaneous quantification of all competing sites while preserving the simplicity of individual prediction models
2Measurement precision
If the model predicts usage of all PAS simultaneously, then the accuracy and completeness of APA regulation understanding is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the complex multi-site prediction problem into multiple simpler binary classification problems, where each model predicts the usage probability of one PAS against others. This segmentation maintains high measurement precision for each site while reducing the computational burden compared to a single monolithic model
Solution Approach 2:
The patent changes the output parameters from binary predictions to normalized usage probabilities that sum to 1 across all PAS. This parameter transformation enables simultaneous prediction of all sites with accurate quantification while the computational complexity remains manageable due to the simplicity of individual binary classification models
3Productivity
If the system processes multiple PAS simultaneously, then the efficiency and productivity of APA analysis is improved, but the model architecture becomes more complex
Solution Approach 1:
The patent segments the model architecture into multiple independent binary classification models, each handling one PAS. This segmentation enables simultaneous processing of multiple PAS, improving productivity through parallel computation while keeping each individual model architecture simple and manageable
Solution Approach 2:
The patent creates a universal framework where the same binary classification model structure can be applied to predict any PAS usage probability. This multi-functionality allows efficient simultaneous analysis of multiple PAS without requiring different specialized models for each site, thereby improving productivity without proportionally increasing architectural complexity
Data Source
AI summary
A method for calculating usage of all alternative polyadenylation sites (PAS) in a genomic sequence includes receiving plural genomic sub-sequences centered on corresponding PAS; processing each genomic sub-sequence of the plural genomic sequences, with a corresponding neural network of plural neural networks; supplying plural outputs of the plural neural networks to an interaction layer that includes plural forward Bidirectional Long Short Term Memory Network (Bi-LSTM) cells and plural backward Bi-LSTM cells, wherein each pair of a forward Bi-LSTM cell and a backward Bi-LSTM cell uniquely receives a corresponding output, of the plural outputs, from a corresponding neural network; and generating a scalar value for each PAS, based on an output from a corresponding pair of the forward Bi-LSTM cell and the backward Bi-LSTM cell.