In recommendation systems, the timely delivery of new content to their relevant audiences is critical for generating a growing and high quality collection of content for all users. The nature of this problem requires retrieval models to be able to make inferences in real time and with high relevance. There are two specific challenges for cold start contents. First, the information loss problem in a standard Two Tower model, due to the limited feature interactions between the user and item towers, is exacerbated for cold start items due to training data sparsity. Second, the huge volume of user-generated content in industry applications today poses a big bottleneck in the end-to-end latency of recommending new content. To overcome the two challenges, we propose a novel architecture, the Item History Model (IHM). IHM directly injects user-interaction information into the item tower to overcome information loss. In addition, IHM incorporates an inductive structure using attention-based pooling to eliminate the need for recurring training, a key bottleneck for the real-timeness. On both public and industry datasets, we demonstrate that IHM can not only outperform baselines in recommending cold start contents, but also achieves SoTA real-timeness in industry applications.

MOTIVATION: Genome-wide association studies revealed that most disease-associated single nucleotide polymorphisms (SNPs) are located in regulatory regions within introns or in regions between genes. Regulatory SNPs (rSNPs) are such SNPs that affect gene regulation by changing transcription factor (TF) binding affinities to genomic sequences. Identifying potential rSNPs is crucial for understanding disease mechanisms. In silico methods that evaluate the impact of SNPs on TF binding affinities are not scalable for large-scale analysis. RESULTS: We describe A: ffinity T: esting for regulatory SNP: s (atSNP), a computationally efficient R package for identifying rSNPs in silico. atSNP implements an importance sampling algorithm coupled with a first-order Markov model for the background nucleotide sequences to test the significance of affinity scores and SNP-driven changes in these scores. Application of atSNP with >20 K SNPs indicates that atSNP is the only available tool for such a large-scale task. atSNP provides user-friendly output in the form of both tables and composite logo plots for visualizing SNP-motif interactions. Evaluations of atSNP with known rSNP-TF interactions indicate that SNP is able to prioritize motifs for a given set of SNPs with high accuracy. AVAILABILITY AND IMPLEMENTATION: https://github.com/keleslab/atSNP. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: ChIP-seq technology enables investigators to study genome-wide binding of transcription factors and mapping of epigenomic marks. Although the availability of basic analysis tools for ChIP-seq data is rapidly increasing, there has not been much progress on the related design issues. A challenging question for designing a ChIP-seq experiment is how deeply should the ChIP and the control samples be sequenced? The answer depends on multiple factors some of which can be set by the experimenter based on pilot/preliminary data. The sequencing depth of a ChIP-seq experiment is one of the key factors that determine whether all the underlying targets (e.g. binding locations or epigenomic profiles) can be identified with a targeted power. RESULTS: We developed a statistical framework named CSSP (ChIP-seq Statistical Power) for power calculations in ChIP-seq experiments by considering a local Poisson model, which is commonly adopted by many peak callers. Evaluations with simulations and data-driven computational experiments demonstrate that this framework can reliably estimate the power of a ChIP-seq experiment at different sequencing depths based on pilot data. Furthermore, it provides an analytical approach for calculating the required depth for a targeted power while controlling the false discovery rate at a user-specified level. Hence, our results enable researchers to use their own or publicly available data for determining required sequencing depths of their ChIP-seq experiments and potentially make better use of the multiplexing functionality of the sequencers. Evaluation of power for multiple public ChIP-seq datasets indicate that, currently, typical ChIP-seq studies are powered well for detecting large fold changes of ChIP enrichment over the control sample, but they have considerably less power for detecting smaller fold changes. AVAILABILITY: Available at www.stat.wisc.edu/~zuo/CSSP. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.