Neural networks with circular filters enable data efficient inference of sequence motifs.

MOTIVATION: Nucleic acids and proteins often have localized sequence motifs that enable highly specific interactions. Due to the biological relevance of sequence motifs, numerous inference methods have been developed. Recently, convolutional neural networks (CNNs) have achieved state of the art performance. These methods were able to learn transcription factor binding sites from ChIP-seq data, resulting in accurate predictions on test data. However, CNNs typically distribute learned motifs across multiple filters, making them difficult to interpret. Furthermore, networks trained on small datasets often do not generalize well to new sequences. RESULTS: Here we present circular filters, a novel convolutional architecture, that convolves sequences with circularly permutated variants of the same filter. We motivate circular filters by the observation that CNNs frequently learn filters that correspond to shifted and truncated variants of the true motif. Circular filters enable learning of full-length motifs and allow easy interpretation of the learned filters. We show that circular filters improve motif inference performance over a wide range of hyperparameters as well as sequence length. Furthermore, we show that CNNs with circular filters in most cases outperform conventional CNNs at inferring DNA binding sites from ChIP-seq data. AVAILABILITY AND IMPLEMENTATION: Code is available at https://github.com/christopherblum. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Endoplasmic reticulum aminopeptidase-1 alleles associated with increased risk of ankylosing spondylitis reduce HLA-B27 mediated presentation of multiple antigens.

Ankylosing spondylitis (AS) is a chronic systemic arthritic disease that leads to significant disability and loss of quality of life in the ∼0.5% of the worldwide human population it affects. There is currently no cure for AS and mechanisms underlying its pathogenesis remain unclear. AS is highly genetic, with over 70% of the genetic risk being associated with the presence of HLA-B27 and endoplasmic reticulum aminopeptidase-1 (ERAP1) alleles. Furthermore, gene-gene interactions between HLA-B27 and ERAP1 AS risk alleles have recently been confirmed. Here, we demonstrate that various ERAP1 alleles can differentially mediate surface expression of antigens presented by HLA-B27 on human cells. Specifically, for all peptides tested, we found that an ERAP1 variant containing high AS risk SNPs reduced the amount of the peptide presented by HLA-B27, relative to low AS risk ERAP1 variants. These results were further validated using peptide catalysis assays in vitro, suggesting that high AS risk alleles have an enhanced catalytic activity that more rapidly destroys many HLA-B27-destined peptides, a result that correlated with decreased HLA-B27 presentation of the same peptides. These findings suggest that one mechanism underlying AS pathogenesis may involve an altered ability for AS patients harboring both HLA-B27 and high AS risk ERAP1 alleles to correctly display a variety of peptides to the adaptive arm of the immune system, potentially exposing such individuals to higher AS risk due to abnormal display of pathogen or self-derived peptides by the adaptive immune system.