Recognition of short linear motifs (SLiMs) or peptides by proteins is an important component of many cellular processes. However, due to limited and degenerate binding motifs, prediction of cellular targets is challenging. In addition, many of these interactions are transient and of relatively low affinity. Here, we focus on one of the largest families of SLiM-binding domains in the human proteome, the PDZ domain. These domains bind the extreme C-terminus of target proteins, and are involved in many signaling and trafficking pathways. To predict endogenous targets of PDZ domains, we developed MotifAnalyzer-PDZ, a program that filters and compares all motif-satisfying sequences in any publicly available proteome. This approach enables us to determine possible PDZ binding targets in humans and other organisms. Using this program, we predicted and biochemically tested novel human PDZ targets by looking for strong sequence conservation in evolution. We also identified three C-terminal sequences in choanoflagellates that bind a choanoflagellate PDZ domain, the Monsiga brevicollis SHANK1 PDZ domain (mbSHANK1), with endogenously-relevant affinities, despite a lack of conservation with the targets of a homologous human PDZ domain, SHANK1. All three are predicted to be signaling proteins, with strong sequence homology to cytosolic and receptor tyrosine kinases. Finally, we analyzed and compared the positional amino acid enrichments in PDZ motifsatisfying sequences from over a dozen organisms. Overall, MotifAnalyzer-PDZ is a versatile program to investigate potential PDZ interactions. This proof-ofconcept work is poised to enable similar types of analyses for other SLiM-binding domains (e.g., MotifAnalyzer-Kinase). MotifAnalyzer-PDZ is available at

No abstract

With the increasing use of machine-learning driven algorithmic judgements, it is critical to develop models that are robust to evolving or manipulated inputs. We propose an extensive analysis of model robustness against linguistic variation in the setting of deceptive news detection, an important task in the context of misinformation spread online. We consider two prediction tasks and compare three state-of-theart embeddings to highlight consistent trends in model performance, high confidence misclassifications, and high impact failures. By measuring the effectiveness of adversarial defense strategies and evaluating model susceptibility to adversarial attacks using characterand word-perturbed text, we find that character or mixed ensemble models are the most effective defenses and that character perturbationbased attack tactics are more successful.

No abstract

Machine learning models that learn from dynamic graphs face nontrivial challenges in learning and inference as both nodes and edges change over time. The existing large-scale graph benchmark datasets that are widely used by the community primarily focus on homogeneous node and edge attributes and are static. In this work, we present a variety of large scale, dynamic heterogeneous academic graphs to test the effectiveness of models developed for multi-step graph forecasting tasks. Our novel datasets cover both context and content information extracted from scientific publications across two communities -Artificial Intelligence (AI) and Nuclear Nonproliferation (NN). In addition, we propose a systematic approach to improve the existing evaluation procedures used in the graph forecasting models.