DataTalks.Club Podcast Wiki

Wiki

Bioinformatics Data Science

Bioinformatics data science applies exploration, modeling, network analysis, and open-source software workflows to biological data from sequencing, metagenomics, and multi-omics experiments.

Related Wiki Pages

Data Science Machine Learning Open Source Reproducibility Data Pipelines

Bioinformatics data science is data science applied to biological data. In Bioinformatics Workflows in Practice, Sebastian Ayala Ruano defines the work as taking information generated in the lab. Bioinformatics then uses exploration, analysis, and modeling to interpret results and make predictions around biological systems.

The episode frames bioinformatics as a bridge between experimental biology and computational work. It covers sequencing-to-analysis workflows, metagenomic samples, and abundance tables. It also covers microbial association networks, knowledge graphs, and scientific reporting. Visualization and open-source package ecosystems make those workflows usable beyond one lab.

Wet Lab and Dry Lab Boundaries

The boundary starts with where the data comes from. At 6:27 in Bioinformatics Workflows in Practice, Sebastian Ayala Ruano contrasts biotechnology experiments with bioinformatics analysis. Lab teams test vaccines, compounds, or biological mechanisms. Computational analysis can reduce the number of experiments by proposing better candidates for the lab to try.

At 8:23, the same episode distinguishes wet lab and dry lab work directly. Wet lab work means physical experiments with samples, instruments, and test tubes. Dry lab work is computational. The scientist or engineer works with data, pipelines, simulations, and software.

Bioinformatics data science sits near machine learning and data pipelines. Its biological context shapes the data model, validation, and interpretation.

From Sequencing to Analysis

Sequencing is the input side of many bioinformatics workflows. In the 12:35 and 15:30 parts of Bioinformatics Workflows in Practice, Sebastian Ayala Ruano explains genomic data as DNA sequences made from four nucleotides. The four are adenine, guanine, cytosine, and thymine. A sequencing workflow reads DNA from an environmental or biological sample. It breaks the DNA into small pieces, decodes the order of nucleotides, assembles the pieces, and compares the result with databases or a reference genome.

That turns a physical sample into analyzable data. The downstream work can look like familiar data science because it includes exploration, matching, and feature interpretation. It also includes modeling and comparison against known references.

The biological stakes change the meaning of the features. A DNA segment may be tied to genes, proteins, or population markers. It may also be tied to health traits or organisms observed in a mixed sample.

Metagenomics and Abundance Tables

Metagenomics is the episode’s clearest pipeline example because it turns raw biological material into analytical tables. At 17:56 in Bioinformatics Workflows in Practice, Sebastian Ayala Ruano describes metagenomics as studying DNA from environmental samples rather than from one organism. Environmental samples may come from a lake, soil, or a wastewater treatment plant. They may contain many microorganisms, so the analysis must decode the mixed DNA and reconstruct the genomes or organism signals inside it.

For Sebastian’s wastewater treatment microbiome project, the working data took the form of abundance tables. At 19:41 and 20:10, he describes rows as microorganisms and columns as samples. Values are counts showing how often each organism appears. Those tables make the workflow recognizable to a data scientist. Datasets from multiple studies are combined, categorized by biome, and analyzed for patterns across samples.

Microbiome Network Inference

The microbiome project moves from tabular analysis into graph analysis. In Bioinformatics Workflows in Practice, Sebastian Ayala Ruano explains at 20:10 that microbial association networks are inferred from co-abundance patterns. If two microorganisms often appear together in similar abundance, the workflow creates a possible association between them.

At 24:31, Sebastian names CC Lasso as the method used in that project to infer potential microorganism interactions. The output includes correlation values between microorganism pairs, thresholding, and positive or negative associations.

He’s careful about interpretation because a positive association can suggest coexistence, while a negative one can suggest that one organism appears when another doesn’t. Geography, sampling, and biological context still matter.

The network then becomes a knowledge graph. Around 36:20 and 38:31, Sebastian describes MCW2 Graph as an open-source knowledge graph where microorganisms are nodes and inferred co-abundance patterns are edges. Extra metadata describes metabolites, biomes, and biological processes. Users can explore the data in a Streamlit app or download raw CSV files. They can also open the graph in Neo4j and run graph algorithms such as clustering or centrality analysis.

Open-Source Computational Biology Tools

The tooling discussion shows why open source is part of the bioinformatics data science workflow rather than an add-on. In Bioinformatics Workflows in Practice, Sebastian Ayala Ruano uses MCW2 Graph and VueGen as examples. He also names VueCore and Viewer. These tools make biological data easier to explore, report, and visualize.

VueGen is the reporting example. At 40:00, Sebastian explains it as a Python package that reads a structured directory. The directory can contain tables, plots, network data, or HTML files. VueGen then generates static documents, presentations, and Streamlit apps.

Under the hood, the tool uses Quarto and Streamlit rather than LLM interpretation. That matters for reproducibility because the report is generated from explicit files, paths, and descriptions. It also uses YAML metadata and renderable project structure.

The package ecosystem also shapes what bioinformatics data scientists build with. At 42:29 and 50:25 in the same episode, Sebastian discusses Bioconda and Bioconductor as biology-oriented package ecosystems. He notes that R still has a larger bioinformatics community, while more work is moving to Python. He also warns that some scientific packages are created by scientists who aren’t trained software developers. Because of that, scaling or adapting them can require reading source code, extending the package, or contacting maintainers.

Use these pages for the neighboring concepts named in the workflow.