UMI-VarCal: a new UMI-based variant caller that efficiently improves low-frequency variant detection in paired-end sequencing NGS libraries.

MOTIVATION: Next-generation sequencing has become the go-to standard method for the detection of single-nucleotide variants in tumor cells. The use of such technologies requires a PCR amplification step and a sequencing step, steps in which artifacts are introduced at very low frequencies. These artifacts are often confused with true low-frequency variants that can be found in tumor cells and cell-free DNA. The recent use of unique molecular identifiers (UMI) in targeted sequencing protocols has offered a trustworthy approach to filter out artefactual variants and accurately call low-frequency variants. However, the integration of UMI analysis in the variant calling process led to developing tools that are significantly slower and more memory consuming than raw-reads-based variant callers. RESULTS: We present UMI-VarCal, a UMI-based variant caller for targeted sequencing data with better sensitivity compared to other variant callers. Being developed with performance in mind, UMI-VarCal stands out from the crowd by being one of the few variant callers that do not rely on SAMtools to do their pileup. Instead, at its core runs an innovative homemade pileup algorithm specifically designed to treat the UMI tags in the reads. After the pileup, a Poisson statistical test is applied at every position to determine if the frequency of the variant is significantly higher than the background error noise. Finally, an analysis of UMI tags is performed, a strand bias and a homopolymer length filter are applied to achieve better accuracy. We illustrate the results obtained using UMI-VarCal through the sequencing of tumor samples and we show how UMI-VarCal is both faster and more sensitive than other publicly available solutions. AVAILABILITY AND IMPLEMENTATION: The entire pipeline is available at https://gitlab.com/vincent-sater/umi-varcal-master under MIT license. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Combining MEDLINE and publisher data to create parallel corpora for the automatic translation of biomedical text.

BACKGROUND: Most of the institutional and research information in the biomedical domain is available in the form of English text. Even in countries where English is an official language, such as the United States, language can be a barrier for accessing biomedical information for non-native speakers. Recent progress in machine translation suggests that this technique could help make English texts accessible to speakers of other languages. However, the lack of adequate specialized corpora needed to train statistical models currently limits the quality of automatic translations in the biomedical domain. RESULTS: We show how a large-sized parallel corpus can automatically be obtained for the biomedical domain, using the MEDLINE database. The corpus generated in this work comprises article titles obtained from MEDLINE and abstract text automatically retrieved from journal websites, which substantially extends the corpora used in previous work. After assessing the quality of the corpus for two language pairs (English/French and English/Spanish) we use the Moses package to train a statistical machine translation model that outperforms previous models for automatic translation of biomedical text. CONCLUSIONS: We have built translation data sets in the biomedical domain that can easily be extended to other languages available in MEDLINE. These sets can successfully be applied to train statistical machine translation models. While further progress should be made by incorporating out-of-domain corpora and domain-specific lexicons, we believe that this work improves the automatic translation of biomedical texts.

Matching health information seekers' queries to medical terms.

BACKGROUND: The Internet is a major source of health information but most seekers are not familiar with medical vocabularies. Hence, their searches fail due to bad query formulation. Several methods have been proposed to improve information retrieval: query expansion, syntactic and semantic techniques or knowledge-based methods. However, it would be useful to clean those queries which are misspelled. In this paper, we propose a simple yet efficient method in order to correct misspellings of queries submitted by health information seekers to a medical online search tool. METHODS: In addition to query normalizations and exact phonetic term matching, we tested two approximate string comparators: the similarity score function of Stoilos and the normalized Levenshtein edit distance. We propose here to combine them to increase the number of matched medical terms in French. We first took a sample of query logs to determine the thresholds and processing times. In the second run, at a greater scale we tested different combinations of query normalizations before or after misspelling correction with the retained thresholds in the first run. RESULTS: According to the total number of suggestions (around 163, the number of the first sample of queries), at a threshold comparator score of 0.3, the normalized Levenshtein edit distance gave the highest F-Measure (88.15%) and at a threshold comparator score of 0.7, the Stoilos function gave the highest F-Measure (84.31%). By combining Levenshtein and Stoilos, the highest F-Measure (80.28%) is obtained with 0.2 and 0.7 thresholds respectively. However, queries are composed by several terms that may be combination of medical terms. The process of query normalization and segmentation is thus required. The highest F-Measure (64.18%) is obtained when this process is realized before spelling-correction. CONCLUSIONS: Despite the widely known high performance of the normalized edit distance of Levenshtein, we show in this paper that its combination with the Stoilos algorithm improved the results for misspelling correction of user queries. Accuracy is improved by combining spelling, phoneme-based information and string normalizations and segmentations into medical terms. These encouraging results have enabled the integration of this method into two projects funded by the French National Research Agency-Technologies for Health Care. The first aims to facilitate the coding process of clinical free texts contained in Electronic Health Records and discharge summaries, whereas the second aims at improving information retrieval through Electronic Health Records.

EVA: Exome Variation Analyzer, an efficient and versatile tool for filtering strategies in medical genomics.

BACKGROUND: Whole exome sequencing (WES) has become the strategy of choice to identify a coding allelic variant for a rare human monogenic disorder. This approach is a revolution in medical genetics history, impacting both fundamental research, and diagnostic methods leading to personalized medicine. A plethora of efficient algorithms has been developed to ensure the variant discovery. They generally lead to ~20,000 variations that have to be narrow down to find the potential pathogenic allelic variant(s) and the affected gene(s). For this purpose, commonly adopted procedures which implicate various filtering strategies have emerged: exclusion of common variations, type of the allelics variants, pathogenicity effect prediction, modes of inheritance and multiple individuals for exome comparison. To deal with the expansion of WES in medical genomics individual laboratories, new convivial and versatile software tools have to implement these filtering steps. Non-programmer biologists have to be autonomous combining themselves different filtering criteria and conduct a personal strategy depending on their assumptions and study design. RESULTS: We describe EVA (Exome Variation Analyzer), a user-friendly web-interfaced software dedicated to the filtering strategies for medical WES. Thanks to different modules, EVA (i) integrates and stores annotated exome variation data as strictly confidential to the project owner, (ii) allows to combine the main filters dealing with common variations, molecular types, inheritance mode and multiple samples, (iii) offers the browsing of annotated data and filtered results in various interactive tables, graphical visualizations and statistical charts, (iv) and finally offers export files and cross-links to external useful databases and softwares for further prioritization of the small subset of sorted candidate variations and genes. We report a demonstrative case study that allowed to identify a new candidate gene related to a rare form of Alzheimer disease. CONCLUSIONS: EVA is developed to be a user-friendly, versatile, and efficient-filtering assisting software for WES. It constitutes a platform for data storage and for drastic screening of clinical relevant genetics variations by non-programmer geneticists. Thereby, it provides a response to new needs at the expanding era of medical genomics investigated by WES for both fundamental research and clinical diagnostics.

UMI-Varcal: A Low-Frequency Variant Caller for UMI-Tagged Paired-End Sequencing Data.

The rapid transition from traditional sequencing methods to Next-Generation Sequencing (NGS) has allowed for a faster and more accurate detection of somatic variants (Single-Nucleotide Variant (SNV) and Copy Number Variation (CNV)) in tumor cells. NGS technologies require a succession of steps during which false variants can be silently added at low frequencies. Filtering these artifacts can be a rather difficult task especially when the experiments are designed to look for very low frequency variants. Recently, adding unique molecular barcodes called UMI (Unique Molecular Identifier) to the DNA fragments appears to be a very effective strategy to specifically filter out false variants from the variant calling results (Kukita et al. DNA Res 22(4):269-277, 2015; Newman et al. Nat Biotechnol 34(5):547-555, 2016; Schmitt et al. Proc Natl Acad Sci U S A 109(36):14508-14513). Here, we describe UMI-VarCal (Sater et al. Bioinformatics 36:2718-2724, 2020), which can use the UMI information from UMI-tagged reads to offer a faster and more accurate variant calling analysis.

UMI-Gen: A UMI-based read simulator for variant calling evaluation in paired-end sequencing NGS libraries.

MOTIVATION: With Next Generation Sequencing becoming more affordable every year, NGS technologies asserted themselves as the fastest and most reliable way to detect Single Nucleotide Variants (SNV) and Copy Number Variations (CNV) in cancer patients. These technologies can be used to sequence DNA at very high depths thus allowing to detect abnormalities in tumor cells with very low frequencies. Multiple variant callers are publicly available and are usually efficient at calling out variants. However, when frequencies begin to drop under 1%, the specificity of these tools suffers greatly as true variants at very low frequencies can be easily confused with sequencing or PCR artifacts. The recent use of Unique Molecular Identifiers (UMI) in NGS experiments has offered a way to accurately separate true variants from artifacts. UMI-based variant callers are slowly replacing raw-read based variant callers as the standard method for an accurate detection of variants at very low frequencies. However, benchmarking done in the tools publication are usually realized on real biological data in which real variants are not known, making it difficult to assess their accuracy. RESULTS: We present UMI-Gen, a UMI-based read simulator for targeted sequencing paired-end data. UMI-Gen generates reference reads covering the targeted regions at a user customizable depth. After that, using a number of control files, it estimates the background error rate at each position and then modifies the generated reads to mimic real biological data. Finally, it will insert real variants in the reads from a list provided by the user. AVAILABILITY: The entire pipeline is available at https://gitlab.com/vincent-sater/umigen under MIT license.