A conditional sequence memory method and system for crop complex trait prediction and candidate variant trait contribution scoring

Abstract

The application discloses a conditional sequence memory method for crop complex trait prediction and candidate variation site trait contribution score, which introduces a conditional sequence memory branch triggered by a DNA token sequence condition in a genome long sequence main model, so that the genome long sequence model after connecting the conditional sequence memory branch obtains an explicit memory search, context-related gate control and residual fusion mode for local k-mer sequence characteristics in addition to continuous hidden representation, and serves as a sequence representation basis for crop complex trait prediction, candidate variation site trait contribution score and candidate molecular marker screening. The application can solve the technical problem that the existing method based on the Transformer architecture lacks an explicit addressable local k-mer sequence memory structure and is difficult to explicitly store and directly call repeated local sequence patterns.