Nicolas Fiorini

Deep Learning for Biomedical Information Retrieval: Learning Textual Relevance from Click Logs.

We describe a Deep Learning approach to modeling the relevance of a document’s text to a query, applied to biomedical literature. Instead of mapping each document and query to a common semantic space, we compute a variable-length difference vector between the query and document which is then passed through a deep convolution stage followed by a deep regression network to produce the estimated probability of the document’s relevance to the query. Despite the small amount of training data, this approach produces a more robust predictor than computing similarities between semantic vector representations of the query and document, and also results in significant improvements over traditional IR text factors. In the future, we plan to explore its application in improving PubMed search.

USI: a fast and accurate approach for conceptual document annotation

BackgroundSemantic approaches such as concept-based information retrieval rely on a corpus in which resources are indexed by concepts belonging to a domain ontology. In order to keep such applications up-to-date, new entities need to be frequently annotated to enrich the corpus. However, this task is time-consuming and requires a high-level of expertise in both the domain and the related ontology. Different strategies have thus been proposed to ease this indexing process, each one taking advantage from the features of the document.ResultsIn this paper we present USI (User-oriented Semantic Indexer), a fast and intuitive method for indexing tasks. We introduce a solution to suggest a conceptual annotation for new entities based on related already indexed documents. Our results, compared to those obtained by previous authors using the MeSH thesaurus and a dataset of biomedical papers, show that the method surpasses text-specific methods in terms of both quality and speed. Evaluations are done via usual metrics and semantic similarity.ConclusionsBy only relying on neighbor documents, the User-oriented Semantic Indexer does not need a representative learning set. Yet, it provides better results than the other approaches by giving a consistent annotation scored with a global criterion — instead of one score per concept.

CompPhy: a web-based collaborative platform for comparing phylogenies

BackgroundCollaborative tools are of great help in conducting projects involving distant workers. Recent web technologies have helped to build such tools for jointly editing office documents and scientific data, yet none are available for handling phylogenies. Though a large number of studies and projects in evolutionary biology and systematics involve collaborations between scientists of different institutes, current tree comparison visualization software and websites are directed toward single-user access. Moreover, tree comparison functionalities are dispersed between different software that mainly focus on high level single tree visualization but to the detriment of basic tree comparison features.ResultsThe web platform presented here, named CompPhy, intends to fill this gap by allowing collaborative work on phylogenies and by gathering simple advanced tools dedicated to tree comparison. It offers functionalities for tree edition, tree comparison, supertree inference and data management in a collaborative environment. The latter aspect is a specific feature of the platform, allowing people located in different places to work together at the same time on a common project. CompPhy thus proposes shared tree visualization, both synchronous and asynchronous tree manipulation, data exchange/storage, as well as facilities to keep track of the progress of analyses in working sessions. Specific advanced comparison tools are also available, such as consensus and supertree inference, or automated branch swaps of compared trees. As projects can be readily created and shared, CompPhy is also a tool that can be used easily to interact with students in a educational setting, either in the classroom or for assignments.ConclusionsCompPhy is the first web platform devoted to the comparison of phylogenetic trees allowing real-time distant collaboration on a phylogenetic/phylogenomic project. This application can be accessed freely with a recent browser at the following page of the ATGC bioinformatics platform: http://www.atgc-montpellier.fr/compphy/.

Cutting Edge: Towards PubMed 2.0

Staff from the National Center for Biotechnology Information in the US describe recent improvements to the PubMed search engine and outline plans for the future, including a new experimental site called PubMed Labs.

Coping with Imprecision During a Semi-automatic Conceptual Indexing Process

Concept-based information retrieval is known to be a powerful and reliable process. It relies on a semantically annotated corpus, i.e. resources indexed by concepts organized within a domain ontology. The conception and enlargement of such index is a tedious task, which is often a bottleneck due to the lack of (semi-)automated solutions. In this paper, we first introduce a solution to assist experts during the indexing process thanks to semantic annotation propagation. The idea is to let them position the new resource on a semantic map, containing already indexed resources and to propose an indexation of this new resource based on those of its neighbors. To further help users, we then introduce indicators to estimate the robustness of the indexation with respect to the indicated position and to the annotation homogeneity of nearby resources. By computing these values before any interaction, it is possible to visually inform users on their margins of error, therefore reducing the risk of having a non-optimal, thus unsatisfying, annotation.

A Fast Deep Learning Model for Textual Relevance in Biomedical Information Retrieval

Publications in the life sciences are characterized by a large technical vocabulary, with many lexical and semantic variations for expressing the same concept. Towards addressing the problem of relevance in biomedical literature search, we introduce a deep learning model for the relevance of a documents text to a keyword style query. Limited by a relatively small amount of training data, the model uses pre-trained word embeddings. With these, the model first computes a variable-length Delta matrix between the query and document, representing a difference between the two texts, which is then passed through a deep convolution stage followed by a deep feed-forward network to compute a relevance score. This results in a fast model suitable for use in an online search engine. The model is robust and outperforms comparable state-of-the-art deep learning approaches.

Best Match: New relevance search for PubMed

PubMed is a free search engine for biomedical literature accessed by millions of users from around the world each day. With the rapid growth of biomedical literature—about two articles are added every minute on average—finding and retrieving the most relevant papers for a given query is increasingly challenging. We present Best Match, a new relevance search algorithm for PubMed that leverages the intelligence of our users and cutting-edge machine-learning technology as an alternative to the traditional date sort order. The Best Match algorithm is trained with past user searches with dozens of relevance-ranking signals (factors), the most important being the past usage of an article, publication date, relevance score, and type of article. This new algorithm demonstrates state-of-the-art retrieval performance in benchmarking experiments as well as an improved user experience in real-world testing (over 20% increase in user click-through rate). Since its deployment in June 2017, we have observed a significant increase (60%) in PubMed searches with relevance sort order: it now assists millions of PubMed searches each week. In this work, we hope to increase the awareness and transparency of this new relevance sort option for PubMed users, enabling them to retrieve information more effectively.

PubRunner: A light-weight framework for updating text mining results

Biomedical text mining promises to assist biologists in quickly navigating the combined knowledge in their domain. This would allow improved understanding of the complex interactions within biological systems and faster hypothesis generation. New biomedical research articles are published daily and text mining tools are only as good as the corpus from which they work. Many text mining tools are underused because their results are static and do not reflect the constantly expanding knowledge in the field. In order for biomedical text mining to become an indispensable tool used by researchers, this problem must be addressed. To this end, we present PubRunner, a framework for regularly running text mining tools on the latest publications. PubRunner is lightweight, simple to use, and can be integrated with an existing text mining tool. The workflow involves downloading the latest abstracts from PubMed, executing a user-defined tool, pushing the resulting data to a public FTP, and publicizing the location of these results on the public PubRunner website. This shows a proof of concept that we hope will encourage text mining developers to build tools that truly will aid biologists in exploring the latest publications.

PubMed Labs: an experimental system for improving biomedical literature search

Abstract PubMed is a freely accessible system for searching the biomedical literature, with ∼2.5 million users worldwide on an average workday. In order to better meet our users’ needs in an era of information overload, we have recently developed PubMed Labs (www.pubmed.gov/labs), an experimental system for users to test new search features/tools (e.g. Best Match) and provide feedback, which enables us to make more informed decisions about potential changes to improve the search quality and overall usability of PubMed. In addition, PubMed Labs features a mobile-first and responsive layout that offers better support for accessing PubMed from increasingly popular mobiles and small-screen devices. In this paper, we detail PubMed Labs, its purpose, new features and best practices. We also encourage users to share their experience with us; based on which we are continuously improving PubMed Labs with more advanced features and better user experience.

Towards a Non-oriented Approach for the Evaluation of Odor Quality

When evaluating an odor, non-specialists generally provide descriptions as bags of terms. Nevertheless, these evaluations cannot be processed by classical odor analysis methods that have been designed for trained evaluators having an excellent mastery of professional controlled vocabulary. Indeed, currently, mainly oriented approaches based on learning vocabularies are used. These approaches too restrictively limit the possible descriptors available for an uninitiated public and therefore require a costly learning phase of the vocabulary. The objective of this work is to merge the information expressed by these free descriptions (terms) into a set of non-ambiguous descriptors best characterizing the odor; this will make it possible to evaluate the odors based on non-specialist descriptions. This paper discusses a non-oriented approach based on Natural Language Processing and Knowledge Representation techniques - it does not require learning a lexical field and can therefore be used to evaluate odors with non-specialist evaluators.