EgoCom: A Multi-Person Multi-Modal Egocentric Communications Dataset.

Multi-modal datasets in artificial intelligence (AI) often capture a third-person perspective, but our embodied human intelligence evolved with sensory input from the egocentric, first-person perspective. Towards embodied AI, we introduce the Egocentric Communications (EgoCom) dataset to advance the state-of-the-art in conversational AI, natural language, audio speech analysis, computer vision, and machine learning. EgoCom is a first-of-its-kind natural conversations dataset containing multi-modal human communication data captured simultaneously from the participants' egocentric perspectives. EgoCom includes 38.5 hours of synchronized embodied stereo audio, egocentric video with 240,000 ground-truth, time-stamped word-level transcriptions and speaker labels from 34 diverse speakers. We study baseline performance on two novel applications that benefit from embodied data: (1) predicting turn-taking in conversations and (2) multi-speaker transcription. For (1), we investigate Bayesian baselines to predict turn-taking within 5 percent of human performance. For (2), we use simultaneous egocentric capture to combine Google speech-to-text outputs, improving global transcription by 79 percent relative to a single perspective. Both applications exploit EgoCom's synchronous multi-perspective data to augment performance of embodied AI tasks.

Dependence on the Pyrimidine Biosynthetic Enzyme DHODH Is a Synthetic Lethal Vulnerability in Mutant KRAS-Driven Cancers.

Activating KRAS mutations are major oncogenic drivers in multiple tumor types. Synthetic lethal screens have previously been used to identify targets critical for the survival of KRAS mutant cells, but their application to drug discovery has proven challenging, possibly due in part to a failure of monolayer cultures to model tumor biology. Here, we report the results of a high-throughput synthetic lethal screen for small molecules that selectively inhibit the growth of KRAS mutant cell lines in soft agar. Chemoproteomic profiling identifies the target of the most KRAS-selective chemical series as dihydroorotate dehydrogenase (DHODH). DHODH inhibition is shown to perturb multiple metabolic pathways. In vivo preclinical studies demonstrate strong antitumor activity upon DHODH inhibition in a pancreatic tumor xenograft model.

Identification of the endosomal sorting complex required for transport-I (ESCRT-I) as an important modulator of anti-miR uptake by cancer cells.

Mechanisms of unassisted delivery of RNA therapeutics, including inhibitors of microRNAs, remain poorly understood. We observed that the hepatocellular carcinoma cell line SKHEP1 retains productive free uptake of a miR-21 inhibitor (anti-miR-21). Uptake of anti-miR-21, but not a mismatch (MM) control, induces expression of known miR-21 targets (DDAH1, ANKRD46) and leads to dose-dependent inhibition of cell growth. To elucidate mechanisms of SKHEP1 sensitivity to anti-miR-21, we conducted an unbiased shRNA screen that revealed tumor susceptibility gene 101 (TSG101), a component of the endosomal sorting complex required for transport (ESCRT-I), as an important determinant of anti-proliferative effects of anti-miR-21. RNA interference-mediated knockdown of TSG101 and another ESCRT-I protein, VPS28, improved uptake of anti-miR-21 in parental SKHEP1 cells and restored productive uptake to SKHEP1 clones with acquired resistance to anti-miR-21. Depletion of ESCRT-I in several additional cancer cell lines with inherently poor uptake resulted in improved activity of anti-miR-21. Finally, knockdown of TSG101 increased uptake of anti-miR-21 by cancer cells in vivo following systemic delivery. Collectively, these data support an important role for the ESCRT-I complex in the regulation of productive free uptake of anti-miRs and reveal potential avenues for improving oligonucleotide free uptake by cancer cells.