Pergola: Boosting Visualization and Analysis of Longitudinal Data by Unlocking Genomic Analysis Tools

摘要

class="head no_bottom_margin" id="sec1title">IntroductionFine-grained longitudinal recordings are one of the fastest growing collections of biological and societal data (, ). Behavior is a prime target for these types of measurements () since it constitutes a high-level phenotype that links genetics, development, neurobiology, evolution, environment, and social influences (). Novel high-throughput platforms for monitoring behavior have recently enabled unprecedented amounts of data to be collected (). However, these data present novel challenges, and their effective comparison and reproducible analysis across different systems and platforms will require improved interoperability (, ). Here, we show how standards established for genomics can be repurposed to handle behavioral data in a fully interoperable fashion. This simple solution made it possible to analyze behavioral neuroscience datasets gathered on three model systems (, , ) using standard genomic formats and software tools.High-throughput behavior-monitoring platforms can be classified into two categories: sensor- and video-based systems. Sensor-based systems measure selected parameters such as vocalizations, activity, feeding, and oxygen consumption. For mice, such devices include PHECOMP (), PhenoMaster () and CLAMS (). Video-based systems use computer vision techniques to track movements. Optimized video setups have been developed for the main model organisms, including mouse (), worm (), fly (href="#bib39" rid="bib39" class=" bibr popnode">Robie et al., 2017), and zebrafish (href="#bib33" rid="bib33" class=" bibr popnode">Pérez-Escudero et al., 2014). Even though the recorded signals are different across systems, the outputs are comparable in the sense that they consist of discrete or continuous time series of behavioral quantitative or qualitative readouts. These time series are usually processed using commercial software shipped with the platform, often in combination with ad hoc implemented analyses. Although this approach is entirely suitable for establishing key biological results, the lack of common standards for longitudinal data processing hampers the ability to share established computational analysis procedures and data (href="#bib47" rid="bib47" class=" bibr popnode">Wilkinson et al., 2016), thus potentially limiting comparisons and reproducibility across different systems. Improving interoperability is not, however, an easy task, because it requires the whole community to agree on common standards, formats, and procedures. In genomics, the establishment of such standards has taken about 15 years and work is still in progress (href="#bib17" rid="bib17" class=" bibr popnode">Field et al., 2011, href="#bib28" rid="bib28" class=" bibr popnode">Karsch-Mizrachi et al., 2018). In this work, we show how mature standards can be recycled across research fields to save development time. To the best of our knowledge we document here the first case of a standard repurposing procedure between genomics and behavioral neuroscience.The rationale for this work was our original observation that the storage of genomics data and the way these data are subsequently analyzed supports all the requirements of longitudinal data since both data types share a sequential structure. In a genomic sequence, nucleotides are ordered as a discrete series to which various layers of qualitative and quantitative annotation can be attached. This data structure, which is essentially a large table with a position in each line and various attributes in each column, can hold any sequentially organized information, including longitudinal recordings. By simply treating each nucleotide coordinate as a unit of time, we show that it is possible to use the most common genomic data formats for storing, visualizing, and analyzing behavioral information (including metadata) in a lossless fashion. What makes this data structure attractive for the handling of longitudinal information is not its specification but rather its ubiquitous use in genomics. In this field, the early adoption of strict data storage standards (href="#bib29" rid="bib29" class=" bibr popnode">Kent et al., 2002) has resulted in a huge number of tools built around formats agreed upon and supported by a community of thousands of laboratories around the world. Formats developed to define the position of genomic annotations therefore provide a perfect scaffold to store discrete time series of behavior. In a similar fashion, file formats implemented to represent scores along genomic sequences allow the storage of continuous time series of behavior (href="/pmc/articles/PMC6231116/figure/fig1/" target="figure" class="fig-table-link figpopup" rid-figpopup="fig1" rid-ob="ob-fig1" co-legend-rid="lgnd_fig1">Figure 1). The available genomic tools (href="#bib1" rid="bib1" class=" bibr popnode">Afgan et al., 2018, href="#bib14" rid="bib14" class=" bibr popnode">Ernst and Kellis, 2012, href="#bib22" rid="bib22" class=" bibr popnode">Gentleman et al., 2004, href="#bib36" rid="bib36" class=" bibr popnode">Quinlan and Hall, 2010, href="#bib37" rid="bib37" class=" bibr popnode">Ramírez et al., 2016, href="#bib40" rid="bib40" class=" bibr popnode">Robinson and Thorvaldsdóttir, 2011, href="#bib51" rid="bib51" class=" bibr popnode">Zerbino et al., 2014) are also suitable to deal with time series because they allow both sophisticated multi-scale visualization, genomic arithmetic operations required to conditionally extract and combine data portions, and complex post-processing techniques such as hidden Markov model (HMM) analysis.href="/pmc/articles/PMC6231116/figure/fig1/" target="figure" rid-figpopup="fig1" rid-ob="ob-fig1">class="inline_block ts_canvas" href="/core/lw/2.0/html/tileshop_pmc/tileshop_pmc_inline.html?title=Click%20on%20image%20to%20zoom&p=PMC3&id=6231116_gr1.jpg" target="tileshopwindow">target="object" href="/pmc/articles/PMC6231116/figure/fig1/?report=objectonly">Open in a separate windowclass="figpopup" href="/pmc/articles/PMC6231116/figure/fig1/" target="figure" rid-figpopup="fig1" rid-ob="ob-fig1">Figure 1Formatting of Common Behavioral Recordings into BED as an Example of the Pergola Approach to Format Data(A) Typical tabulated (CSV, TSV, or XLSX) behavioral recording with three events (lines 2–4) being coded with an animal tag (id column), start (event_start column) and end time (event_end column), a typology (type_of_event), and a quantitative value (value column).(B) Once mapped onto the Pergola generic ontology, these column labels are translated into a BED file format where the nucleotide relative index positions (columns 2 and 3) are used to specify the start and termination of the event whose nature is coded in the attribute column (#4) and quantified in the quantification column (#6). The BED format also supports specific coloring for the considered event (RGB values in the last column). A single BED file will contain all the events of an individual animal ID. Pergola will generate a FASTA file name chromosome 1 (“chr1” on the file) that is used to map the time points onto the nucleotide positions.(C) Once formatted this way, data can then be displayed and processed using popular genomics tools such as the Integrative Genomic Viewer (IGV).