Extracting Drug-Protein Relation from Literature Using Ensembles of Biomedical Transformers.

Automatic extraction of relations between drugs/chemicals and proteins from ever-growing biomedical literature is required to build up-to-date knowledge bases in biomedicine. To promote the development of automated methods, BioCreative-VII organized a shared task - the DrugProt track, to recognize drug-protein entity relations from PubMed abstracts. We participated in the shared task and leveraged deep learning-based transformer models pre-trained on biomedical data to build ensemble approaches to automatically extract drug-protein relation from biomedical literature. On the main corpora of 10,750 abstracts, our best system obtained an F1-score of 77.60% (ranked 4th among 30 participating teams), and on the large-scale corpus of 2.4M documents, our system achieved micro-averaged F1-score of 77.32% (ranked 2nd among 9 system submissions). This demonstrates the effectiveness of domain-specific transformer models and ensemble approaches for automatic relation extraction from biomedical literature.

Biomolecular Condensation: A New Phase in Cancer Research.

Multicellularity was a watershed development in evolution. However, it also meant that individual cells could escape regulatory mechanisms that restrict proliferation at a severe cost to the organism: cancer. From the standpoint of cellular organization, evolutionary complexity scales to organize different molecules within the intracellular milieu. The recent realization that many biomolecules can "phase-separate" into membraneless organelles, reorganizing cellular biochemistry in space and time, has led to an explosion of research activity in this area. In this review, we explore mechanistic connections between phase separation and cancer-associated processes and emerging examples of how these become deranged in malignancy. SIGNIFICANCE: One of the fundamental functions of phase separation is to rapidly and dynamically respond to environmental perturbations. Importantly, these changes often lead to alterations in cancer-relevant pathways and processes. This review covers recent advances in the field, including emerging principles and mechanisms of phase separation in cancer.

Exploiting induced vulnerability to overcome PARPi resistance and clonal heterogeneity in BRCA mutant triple-negative inflammatory breast cancer.

Acquired resistance and clonal heterogeneity are critical challenges in cancer treatment, and the lack of effective computational tools hampers the discovery of new treatments to overcome resistance. Using high-throughput transcriptomic databases of compound perturbation profiles, we have developed a bioinformatic strategy for identifying candidate drugs to overcome resistance with combinatorial therapy. We devised this strategy during an investigation into the acquired resistance against PARP inhibitors (PARPi) in a triple-negative inflammatory breast cancer cell line. In this study, we derived multiple PARPi-resistant clones and characterized their transcriptomic adaptations compared to the parental clone. The transcriptomes of the resistant clones showed substantial heterogeneity, highlighting the importance of characterizing multiple clones from the same tumour. Surprisingly, we found that these transcriptomic changes may not actually confer PARPi resistance, but they may nevertheless induce a shared secondary vulnerability. By modeling our data in relation to transcriptomic perturbation profiles of compounds, we uncovered deficiencies in Ras signaling that resulted from transcriptional adaptation to long-term PARPi treatment across multiple resistant clones. Due to these induced deficiencies, we predicted that the resistant clones would be sensitive to pharmacological reinforcement of PARPi-induced transcriptional adaptation. We then experimentally validated this predicted vulnerability that is shared by multiple resistant clones. Our results thus provide a promising paradigm for integrating transcriptomic data with compound perturbation profiles in order to identify drugs that can exploit an induced vulnerability and overcome therapeutic resistance, thus providing another strategy towards precision oncology.

Human Endogenous Retroviruses in Glioblastoma Multiforme.

Glioblastoma multiforme (GBM) is the most aggressive and deadly brain tumor. It is primarily diagnosed in the elderly and has a 5-year survival rate of less than 6% even with the most aggressive therapies. The lack of biomarkers has made the development of immunotherapy for GBM challenging. Human endogenous retroviruses (HERVs) are a group of viruses with long terminal repeat (LTR) elements, which are believed to be relics from ancient viral infections. Recent studies have found that those repetitive elements play important roles in regulating various biological processes. The differentially expressed LTR elements from HERVs are potential biomarkers for immunotherapy to treat GBM. However, the understanding of the LTR element expression in GBM is greatly lacking. METHODS: We obtained 1077.4 GB of sequencing data from public databases. These data were generated from 111 GBM tissue studies, 30 GBM cell lines studies, and 45 normal brain tissues studies. We analyzed repetitive elements that were differentially expressed in GBM and normal brain samples. RESULTS: We found that 48 LTR elements were differentially expressed (p-value < 0.05) between GBM and normal brain tissues, of which 46 were HERV elements. Among these 46 elements, 34 significantly changed HERVs belong to the ERV1 superfamily. Furthermore, 43 out of the 46 differentially expressed HERV elements were upregulated. CONCLUSION: Our results indicate significant differential expression of many HERV LTR elements in GBM and normal brain tissues. Expression levels of these elements could be developed as biomarkers for GBM treatments.

Time-sensitive clinical concept embeddings learned from large electronic health records.

BACKGROUND: Learning distributional representation of clinical concepts (e.g., diseases, drugs, and labs) is an important research area of deep learning in the medical domain. However, many existing relevant methods do not consider temporal dependencies along the longitudinal sequence of a patient's records, which may lead to incorrect selection of contexts. METHODS: To address this issue, we extended three popular concept embedding learning methods: word2vec, positive pointwise mutual information (PPMI) and FastText, to consider time-sensitive information. We then trained them on a large electronic health records (EHR) database containing about 50 million patients to generate concept embeddings and evaluated them for both intrinsic evaluations focusing on concept similarity measure and an extrinsic evaluation to assess the use of generated concept embeddings in the task of predicting disease onset. RESULTS: Our experiments show that embeddings learned from information within one visit (time window zero) improve performance on the concept similarity measure and the FastText algorithm usually had better performance than the other two algorithms. For the predictive modeling task, the optimal result was achieved by word2vec embeddings with a 30-day sliding window. CONCLUSIONS: Considering time constraints are important in training clinical concept embeddings. We expect they can benefit a series of downstream applications.