Determining usefulness of biomedical text mining systems requires realistic task definition and data selection criteria without artificial constraints, measuring performance aspects that go beyond traditional metrics. The BioCreative III Protein-Protein Interaction (PPI) tasks were motivated by such considerations, trying to address aspects including how the end user would oversee the generated output, for instance by providing ranked results, textual evidence for human interpretation or measuring time savings by using automated systems. Detecting articles describing complex biological events like PPIs was addressed in the Article Classification Task (ACT), where participants were asked to implement tools for detecting PPI-describing abstracts. Therefore the BCIII-ACT corpus was provided, which includes a training, development and test set of over 12,000 PPI relevant and non-relevant PubMed abstracts labeled manually by domain experts and recording also the human classification times. The Interaction Method Task (IMT) went beyond abstracts and required mining for associations between more than 3,500 full text articles and interaction detection method ontology concepts that had been applied to detect the PPIs reported in them. A total of 11 teams participated in at least one of the two PPI tasks (10 in ACT and 8 in the IMT) and a total of 62 persons were involved either as participants or in preparing data sets/evaluating these tasks. Per task, each team was allowed to submit five runs offline and another five online via the BioCreative Meta-Server. From the 52 runs submitted for the ACT, the highest Matthew's Correlation Coefficient (MCC) score measured was 0.55 at an accuracy of 89% and the best AUC iP/R was 68%. Most ACT teams explored machine learning methods, some of them also used lexical resources like MeSH terms, PSI-MI concepts or particular lists of verbs and nouns, some integrated NER approaches. For the IMT, a total of 42 runs were evaluated by comparing systems against manually generated annotations done by curators from the BioGRID and MINT databases. The highest AUC iP/R achieved by any run was 53%, the best MCC score 0.55. In case of competitive systems with an acceptable recall (above 35%) the macro-averaged precision ranged between 50% and 80%, with a maximum F-Score of 55%. The results of the ACT task of BioCreative III indicate that classification of large unbalanced article collections reflecting the real class imbalance is still challenging. Nevertheless, text-mining tools that report ranked lists of relevant articles for manual selection can potentially reduce the time needed to identify half of the relevant articles to less than 1/4 of the time when compared to unranked results. Detecting associations between full text articles and interaction detection method PSI-MI terms (IMT) is more difficult than might be anticipated. This is due to the variability of method term mentions, errors resulting from pre-processing of articles provided as PDF files, and the heterogeneity and different granularity of method term concepts encountered in the ontology. However, combining the sophisticated techniques developed by the participants with supporting evidence strings derived from the articles for human interpretation could result in practical modules for biological annotation workflows.

The International Molecular Exchange (IMEx) Consortium provides scientists with a single body of experimentally verified protein interactions curated in rich contextual detail to an internationally agreed standard. In this update to the work of the IMEx Consortium, we discuss how this initiative has been working in practice, how it has ensured database sustainability, and how it is meeting emerging annotation challenges through the introduction of new interactor types and data formats. Additionally, we provide examples of how IMEx data are being used by biomedical researchers and integrated in other bioinformatic tools and resources.

High throughput technologies such as deep sequencing and proteomics are increasingly becoming mainstream in clinical practice and support diagnosis and patient stratification. Developing computational models that recapitulate cell physiology and its perturbations in disease is a required step to help with the interpretation of results of high content experiments and to devise personalized treatments. As complete cell-models are difficult to achieve, given limited experimental information and insurmountable computational problems, approximate approaches should be considered. We present here a general approach to modeling complex diseases by embedding patient-specific genomics data into actionable logic models that take into account prior knowledge. We apply the strategy to acute myeloid leukemia (AML) and assemble a network of logical relationships linking most of the genes that are found frequently mutated in AML patients. We derive Boolean models from this network and we show that by priming the model with genomic data we can infer relevant patient-specific clinical features. Here we propose that the integration of literature-derived causal networks with patient-specific data should be explored to help bedside decisions.

Autism spectrum disorder (ASD) comprises a large group of neurodevelopmental conditions featuring, over a wide range of severity and combinations, a core set of manifestations (restricted sociality, stereotyped behavior and language impairment) alongside various comorbidities. Common and rare variants in several hundreds of genes and regulatory regions have been implicated in the molecular pathogenesis of ASD along a range of causation evidence strength. Despite significant progress in elucidating the impact of few paradigmatic individual loci, such sheer complexity in the genetic architecture underlying ASD as a whole has hampered the identification of convergent actionable hubs hypothesized to relay between the vastness of risk alleles and the core phenotypes. In turn this has limited the development of strategies that can revert or ameliorate this condition, calling for a systems-level approach to probe the cross-talk of cooperating genes in terms of causal interaction networks in order to make convergences experimentally tractable and reveal their clinical actionability. As a first step in this direction, we have captured from the scientific literature information on the causal links between the genes whose variants have been associated with ASD and the whole human proteome. This information has been annotated in a computer readable format in the SIGNOR database and is made freely available in the resource website. To link this information to cell functions and phenotypes, we have developed graph algorithms that estimate the functional distance of any protein in the SIGNOR causal interactome to phenotypes and pathways. The main novelty of our approach resides in the possibility to explore the mechanistic links connecting the suggested gene-phenotype relations.

The coronavirus disease 2019 (COVID-19) pandemic has caused more than 2.3 million casualties worldwide and the lack of effective treatments is a major health concern. The development of targeted drugs is held back due to a limited understanding of the molecular mechanisms underlying the perturbation of cell physiology observed after viral infection. Recently, several approaches, aimed at identifying cellular proteins that may contribute to COVID-19 pathology, have been reported. Albeit valuable, this information offers limited mechanistic insight as these efforts have produced long lists of cellular proteins, the majority of which are not annotated to any cellular pathway. We have embarked in a project aimed at bridging this mechanistic gap by developing a new bioinformatic approach to estimate the functional distance between a subset of proteins and a list of pathways. A comprehensive literature search allowed us to annotate, in the SIGNOR 2.0 resource, causal information underlying the main molecular mechanisms through which severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and related coronaviruses affect the host–cell physiology. Next, we developed a new strategy that enabled us to link SARS-CoV-2 interacting proteins to cellular phenotypes via paths of causal relationships. Remarkably, the extensive information about inhibitors of signaling proteins annotated in SIGNOR 2.0 makes it possible to formulate new potential therapeutic strategies. The proposed approach, which is generally applicable, generated a literature-based causal network that can be used as a framework to formulate informed mechanistic hypotheses on COVID-19 etiology and pathology.

We describe a large-scale community effort to build an open-access, interoperable, and computable repository of COVID-19 molecular mechanisms - the COVID-19 Disease Map. We discuss the tools, platforms, and guidelines necessary for the distributed development of its contents by a multi-faceted community of biocurators, domain experts, bioinformaticians, and computational biologists. We highlight the role of relevant databases and text mining approaches in enrichment and validation of the curated mechanisms. We describe the contents of the Map and their relevance to the molecular pathophysiology of COVID-19 and the analytical and computational modelling approaches that can be applied for mechanistic data interpretation and predictions. We conclude by demonstrating concrete applications of our work through several use cases and highlight new testable hypotheses.