121-P: INFORMATIVE PARALLEL SEQUENCING: HAPLOTYPE CLASSIFICATION BY MACHINE LEARNING

Abstract

Aim To validate the transformation of the classic “molecular typing” problem into one of classifying patients and donors by HPA and class 1 HLA immunomolecular signatures for routine procurement of suitable platelets and ultimately stem cells on the basis of epitopes encoded by reconstructed haplotypes. Methods Key elements of the process are: > Informative parallel sequencing & “machine learning - as with classifying consumers by selected attributes, our method classifies patients and donors by immunomolecular signatures derived from the sequence at a small number of “informative” sites; > Haplotype decoding - limiting the sequence determination to a small number of sites substantially reduces the degree of ambiguity inherent in the genotypes: this reduction in complexity permits us to “solve the phase problem” by reconstructing allele and haplotype pairs. > Parallel processing: “pooling” - process performance is further enhanced by simultaneous sequence determination at multiple designated sites for multiple pooled samples: the fig shows reaction patterns for two HLA.1 informative sites for 4 patients in a single tube reaction.[figure1] Results 296 hematologic & non-hematologic oncology patients and 160 candidate platelet donors were classified by HPA and class I HLA profiles: the process achieved 100% concordance with high-resolution HLA types available for 42 patients. In addition, we recovered haplotype pairs. Conclusions Our process, combining informative parallel sequencing with machine learning, haplotype recovery and sample pooling, offers an effective approach to the systematic large-scale classification of recipients and candidate donors by HPA and HLA molecular signatures. Hashmi: BioMolecular Analytics, LLC: Employee. Patel: BioMolecular Analytics, LLC: Employee. Seul: BioMolecular Analytics, LLC: Employee.