Transformer Sequence Models for RNA Editing Specificity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current RNA editing systems face limitations such as aberrant effector activity, delivery barriers, unintended transcriptomic modifications, immunogenicity, and suboptimal tissue targeting, particularly with AAV capsids, and the complexity of regulatory elements makes determining their sequence determinants difficult.
Innovation Solution
Employ machine learning models, specifically transformer-based encoder-decoder architectures, to predict and optimize deamination efficiency and specificity of guide RNAs and AAV capsid proteins by training on nucleic acid sequences and structural features, enabling in silico screening and design of sequences for targeted RNA editing and tissue-specific delivery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If naturally occurring AAV capsids are used for gene therapy delivery, then the system is simple and well-characterized, but the tissue targeting is suboptimal with poor infectivity for specific tissues and competing liver tropism
Solution Approach 1:
The patent applies local quality by making specific localized modifications to the AAV capsid structure, particularly at the fiber interface region (residues 581-589 of VP1). Instead of redesigning the entire capsid, the invention introduces targeted amino acid substitutions at specific positions to alter tissue tropism while maintaining overall capsid integrity and function. This localized modification approach enables improved muscle and CNS targeting while reducing liver tropism.
Solution Approach 2:
The patent employs parameter changes by systematically varying amino acid sequences at specific capsid positions to optimize tissue targeting. The invention tests multiple amino acid substitutions (e.g., L581Q, L581R, L581K, L581E, L581M, L581V, L581I, L581W, L581F, L581Y, L581A) to identify combinations that enhance infectivity for desired tissues while reducing off-target effects. This parameter optimization approach allows precise control over capsid behavior.
2Reliability
If higher doses of AAV are administered to ensure infection of desired tissues, then the infectivity for target tissues improves, but dose-dependent liver toxicity increases
Solution Approach 1:
The patent applies local quality by modifying specific regions of the AAV capsid (fiber interface residues 581-589) to redirect tropism away from the liver. These localized changes alter the capsid's interaction with cellular receptors, enabling preferential binding to muscle and CNS cells while reducing hepatic uptake. This spatially selective modification resolves the contradiction by changing where the virus delivers its payload without increasing overall dose.
Solution Approach 2:
The patent converts the harmful liver tropism into a beneficial feature by engineering capsids with reduced liver affinity. The modified capsids that naturally avoid the liver are then optimized for enhanced muscle and CNS targeting. This approach transforms the problematic off-target accumulation into a design advantage, allowing lower doses to achieve the same therapeutic effect in target tissues without liver toxicity.
3Reliability
If naturally occurring AAV capsids are used, then the system is well-characterized and safe, but immunological memory challenges arise with pre-immune patient populations and repeat dosing
Solution Approach 1:
The patent applies local quality by introducing targeted amino acid substitutions at the fiber interface region of the AAV capsid. These localized changes create capsid variants that are sufficiently different from natural serotypes to evade pre-existing immune responses, yet maintain sufficient similarity to preserve capsid assembly and function. This selective modification strategy enables use in pre-immune patients while retaining the safety advantages of AAV-based delivery.
Solution Approach 2:
The patent creates composite capsid structures by combining elements from different AAV serotypes or introducing non-natural amino acid sequences at specific positions. These chimeric or hybrid capsids possess novel surface properties that reduce recognition by pre-existing neutralizing antibodies while maintaining cellular entry functions. This composite approach expands the applicability to patients with pre-existing immunity to common AAV serotypes.
4Loss of information
If the genome is analyzed to discover sequence determinants of regulatory elements, then comprehensive data is available, but the repetitive nature and multiple functions of the genome make determination difficult
Solution Approach 1:
The patent applies the extraction principle by isolating and analyzing specific regulatory sequences in controlled experimental contexts. Rather than attempting to decipher the entire genome simultaneously, the invention extracts individual enhancer, promoter, or insulator sequences and tests their activity in standardized reporter assays. This extraction approach separates the determination of sequence determinants from the complexity of the full genome, enabling systematic identification of functional elements.
Solution Approach 2:
The patent applies segmentation by dividing regulatory elements into discrete testable units with defined boundaries and functions. The invention analyzes regulatory sequences in modular fashion, testing individual enhancers, promoters, and insulators separately to identify their specific sequence determinants. This segmentation of the genome into functional modules makes the analysis tractable despite the overall complexity and repetitiveness of genomic DNA.
Data Source
AI summary
A model comprising an encoder block and a decoder block are obtained. Nucleic acid sequence information for a scaffold formed between a gRNA and a target RNA including components corresponding to the gRNA and target RNA, and structural information comprising a base-pairing probability matrix for the scaffold, are inputted into the model. The encoder block comprises a first attention mechanism that receives the sequence information and the structural information. The decoder block includes a first sub-portion including a second and third attention mechanism and receives, as input, output generated from the encoder block. Output from the model is received, including predicted metrics for efficiency or specificity of deamination of target nucleotide positions in the target RNA by a deamination enzyme facilitated by hybridization of the gRNA to the target RNA.