Neural network architectures for scoring and visualizing biological sequence variations using molecular phenotype, and systems and methods therefor
Inventors
Delong, Andrew • Frey, Brendan
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11568960-B2
Publication Date
2023-01-31
Expiration Date
Abstract
Systems and methods for scoring and visualizing the effects of variants in biological sequences. Variants may include substitutions, insertions and deletions. The method comprises encoding biological sequences as vector sequences and then operating a neural network in the forward-propagation mode and possibly in the back-propagation mode to compute variant scores. Variant scores are determined by normalizing the gradients. Variant scores may be used to select a subset of variants, which are then used to produce modified vector sequences which are analyzed by the neural network operating in forward-propagation mode, to determine improved variant scores. The variant scores may be visualized using black and white, greyscale or colored elements that are arranged in blocks with dimensions corresponding to different possible symbols and the length of the sequence. These blocks are aligned with the biological sequence, which is illustrated by a symbol sequence arranged in a line.
Core Innovation
The invention provides systems and methods for determining and visualizing variant effects of biological sequence variants, including substitution variants, at one or more variant positions in a reference biological sequence. The reference biological sequence is characterized by a sequence of reference symbols, and each substitution variant is associated with a variant position. Variant biological sequences are obtained as modifications of the reference biological sequence by the set of substitution variants.
A molecular phenotype neural network (MPNN) is used to determine a molecular phenotype for the reference biological sequence and variant molecular phenotypes for the variant biological sequences. The MPNN operates in a forward-propagation mode to output molecular phenotypes that quantify aspects of biological molecules of cells, and the MPNN operates in a back-propagation mode to determine gradients for substitution variants and gradients for the reference symbol at the respective variant positions.
Variant scores are determined by comparing, using a comparator, the gradient for the substitution variant and the gradient for the reference symbol at the respective variant position. The comparison and scoring can be based on gradient subtraction and absolute difference, and variant scores can be normalized using normalization based on max or mean absolute scores, including division by a maximum absolute variant score among substitution variants at an associated variant position.
The disclosed visualization approaches align variant-score tracks with the reference sequence and represent mutation tracks using blocks aligned to sequence positions in grayscale, color, or black-white matrices. Visual elements can be scaled by max score and use a colormap to indicate sign or magnitude of the variant scores. The visualization can include annotations and auxiliary tracks such as conservation, allele frequency, and a ChIP-Seq track.
Claims Coverage
Two independent claims are directed to a method and a system for determining variant scores for substitution variants at one or more positions, using an MPNN with forward-propagation molecular phenotypes and back-propagation gradients that are compared by a comparator.
Variant scoring using MPNN gradients compared to reference symbols
Determining a variant score for a set of substitution variants at one or more variant positions in a reference biological sequence using a molecular phenotype neural network (MPNN) by operating the MPNN in a forward-propagation mode to determine molecular phenotypes and in a back-propagation mode to determine a gradient for each substitution variant and a gradient for the reference symbol at the respective variant position, and determining the variant score at least in part by comparing, using a comparator, the gradient for the substitution variant and the gradient for the reference symbol at the respective variant position.
System modules for MPNN-based gradient comparison
Providing a system comprising a module configured to obtain a reference biological sequence and a set of substitution variants associated with variant positions, an MPNN for determining molecular phenotypes and gradients in forward-propagation and back-propagation modes, and a comparator for determining a variant score for each substitution variant at least in part by comparing the gradient for the substitution variant and the gradient for the reference symbol at the respective variant position.
Both independent claims center on using an MPNN to produce molecular phenotypes in forward-propagation mode and gradients in back-propagation mode, then using a comparator to determine variant scores by comparing each substitution-variant gradient to the corresponding reference-symbol gradient at the same variant position.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?