# Bioinformatics Research Platform

> A no-code, polyglot environment for bioinformatics EDA and ML inference

- HTML version: https://robbiepalmer.me/projects/bioinformatics-platform
- Status: completed
- Started: 2021-01-01
- Product: https://sonraianalytics.com/biomarker-platform-features/
- Technologies: Python, R, Shiny, Plotly

# Vision

The goal was to accelerate all bioinformaticians and democratize data analysis, enabling non-technical domain experts to perform complex analyses that were previously inaccessible without coding skills.

Crucially, this abstraction could not come at the cost of capability. Technical bioinformaticians needed to retain the full power and flexibility of their scripting environments, ensuring they didn't feel restricted by a "dumbed-down" tool.

# Problem Statement

Before this platform, creating new analysis applications was a slow, repetitive process:

* **Slow Time-to-Market**: Building a new app took weeks of dedicated engineering time.
* **Siloed Ecosystems**: Tools were fragmented between R and Python, forcing teams to choose one ecosystem and lose access to niche libraries in the other.
* **Lack of Standards**: There were no consistent CI/CD or DevOps practices, leading to fragile deployments.

# User Stories

* **As a non-technical researcher**, I want to perform exploratory data analysis (EDA) using a GUI so I can find insights without learning R or Python.
* **As a bioinformatician**, I want to leverage specific R packages (like Bioconductor) alongside Python ML libraries in the same workflow.
* **As an engineer**, I want to spin up a new standardized application in days rather than weeks, with built-in CI/CD.

# Platform Capabilities

* **Rapid Application Development**: Consolidated underlying frameworks to reduce new app creation time from weeks to days.
* **Polyglot Support**: Enabled seamless interoperability between R (Shiny) and Python, allowing access to the full breadth of bioinformatics and ML libraries in both ecosystems.
* **No-Code Interface**: Intuitive GUI for complex statistical analysis, machine learning inference, and multi-omic interoperability.
* **Professional DevOps**: Standardized CI/CD pipelines and deployment practices for stability and speed.

# How the platform worked

Researchers used a graphical interface to explore biological datasets, run
statistical analyses, and apply machine learning models without writing code.
Each analysis application was a microservice that controlled its own data,
processing, and visualisation using Python with Plotly or R with Shiny. Apps
could pass data to one another and be combined into a workflow.

These workflows formed a directed acyclic graph, or DAG. An app's output could
feed several downstream apps, with data flowing forward through the analysis.
The diagram shows an example with two branches.

```mermaid
flowchart TB
Researcher["Researcher"]

subgraph Platform["Bioinformatics Research Platform"]
Interface["No-code interface"]
subgraph Workflow["Example analysis workflow · DAG"]
First["Python analysis app<br/>Own data · processing · Plotly"]
Next["R analysis app<br/>Own data · processing · Shiny"]
Inference["Python model inference app<br/>Own data · processing · Plotly"]
First -->|"pass data"| Next
First -->|"pass data"| Inference
end
end

Researcher -->|"configure and run analyses"| Interface
Interface -->|"run workflow"| First
Next -->|"visualised results"| Interface
Inference -->|"visualised predictions"| Interface
```

The researcher configured and ran analyses through the interface.
Each app managed its data, performed its processing, and presented its own
results. This followed a data-mesh approach: ownership stayed with the app,
while apps exchanged data through the workflow instead of relying on a shared
data store.

Shared application frameworks and standardised CI/CD pipelines let engineers
add new analysis tools in days rather than rebuilding each application from
scratch over several weeks.

## Initiatives

- [Personalized Medicine](https://robbiepalmer.me/initiatives/personalized-medicine.md): Connected pathology outputs with genomic and other omic data so researchers could investigate patient-specific signals in one environment.

---

Markdown index of this site: https://robbiepalmer.me/llms.txt
