Supplementary Information

S4. Reagents and algorithms

A methods table. Bench first, then the model.

The article gives the governing equations. This supplement lists what actually went into the flask: wet-lab technique, statistical tests, Bioconductor packages, and the learners used to rank biomarkers.

panel = ar g S max AUROC (f_{S})

(S1)

Equation (S1) is the practical objective of the RA programme: choose a small feature set that still classifies. The tools below are how that set is reached.

Figure S4. PA teaching films of the left hand. The pipelines below are built for the rheumatoid pattern on the right.

Figure S4a. Schematic volcano used when teaching DESeq2 / limma contrasts. Not a published result.

Figure S4b. Schematic ECS heatmap, sex by diagnosis.

Figure S4c. The same career, read as a gel.

S4.1 Wet Lab Techniques

Foundations · Bench skills

Hands-on molecular biology training from undergraduate and graduate coursework: preparing, amplifying, and visualizing nucleic acid samples at the bench before they ever reach a script.

Used. PCR, Gel electrophoresis, DNA/RNA extraction.

S4.2 Programming & Scripting

Foundations · Languages & tooling

The scripting bench everything downstream runs on: R and Python for analysis, SQL for structured data, and Linux/Jupyter for reproducible execution.

Used. R, Python, SQL / NoSQL, Linux, Jupyter notebooks, Command-line tools, pandas / NumPy, BeautifulSoup (web scraping).

S4.3 Statistics & Data Analysis

Core Analysis · Inference

Descriptive and inferential statistics used to validate assumptions and test relationships before any model gets trained, from hypothesis tests to correlation and regression.

Used. Hypothesis testing (t-test, ANOVA), Chi-square / Kolmogorov-Smirnov, Pearson / Spearman correlation, Linear regression, Descriptive statistics, Exploratory data analysis (EDA).

S4.4 Data Science & Visualization

Core Analysis · Communication

Turning cleaned data into decision-ready insight, from client-facing dashboards to publication-style plots.

Used. ggplot2, Matplotlib / Seaborn / Plotly, Tableau / Power BI, Data mining.

S4.5 Genomics & Transcriptomics

Domain Bioinformatics · RNA & sequencing

Differential expression with DESeq2 (moderated fold-change and dispersion estimation) and limma, plus edgeR for count-based modeling, ChIP-seq, and CRISPR guide design for functional follow-up.

Used. RNA-seq, Microarray analysis, DESeq2, limma, edgeR, samtools, ChIP-seq analysis, clusterProfiler (pathway enrichment), CRISPR-Cas9 guide design, Heatmap/volcano/PCA visualization.

Also known. FastQC / MultiQC, STAR / HISAT2, Salmon, CRISPOR / CHOPCHOP.

S4.6 Methylomics & Epigenomics

Domain Bioinformatics · Epigenetics

DNA methylation signal pulled from public GEO series and integrated alongside transcriptomic data to strengthen candidate biomarker panels.

Used. DNA methylation analysis, GEOquery data retrieval, Methylomics preprocessing.

Also known. Bismark, methylKit, minfi.

S4.7 Proteomics & Molecular Docking

Domain Bioinformatics · Structure

Structure-based virtual screening with AutoDock and PyMOL, evaluating how small-molecule candidates bind disease-relevant protein targets.

Used. Molecular docking, AutoDock, PyMOL, Virtual/in-silico screening, Binding-affinity scoring.

Also known. AutoDock Vina, Open Babel.

S4.8 Multi-Omics & Cross-Omics Integration

Domain Bioinformatics · Integration

Combining transcriptomic, methylomic, and genomic layers into one model, including WGCNA co-expression networks and Mendelian randomization for causal inference.

Used. Horizontal/vertical/gene-level integration, WGCNA (co-expression networks), TwoSampleMR (Mendelian randomization), Feature selection (Boruta, LASSO).

Also known. mixOmics, MOFA (Multi-Omics Factor Analysis).

S4.9 Machine Learning & Modeling

Advanced & Applied · Predictive modeling

Supervised and unsupervised modeling for biomarker classification, from classical ensembles to deep learning and generative networks for synthetic data.

Used. Logistic regression, Random forest, XGBoost, SVM / KNN, Gradient boosting, Deep neural networks, CNN (image classification), GANs (TensorFlow/PyTorch), Hyperparameter tuning, SMOTE / class balancing, scikit-learn.

S4.10 Web Development

Advanced & Applied · Applications

Shiny applications that wrap DESeq2/limma pipelines into interactive dashboards, so lab members can explore results without running R scripts themselves.

Used. R Shiny, Interactive dashboards, Web application deployment.

S4.11 AI Model Evaluation & Review Writing

Advanced & Applied · AI research support, Mercor

Learned through hands-on AI-lab evaluation work at Mercor: designing rubrics, judging model-generated bioinformatics analyses, and writing structured, evidence-based reviews of AI outputs.

Used. Rubric and evaluation-criteria design, Model-output review writing, Benchmark task design, Annotation quality assurance, Human-in-the-loop AI workflows.

S4.12 AI Tools & Productivity

Advanced & Applied · Workflow

AI assistants used day to day to move faster, from building this portfolio to code review and research synthesis.

Used. Claude (Claude Code), GitHub Copilot, Perplexity, NotebookLM.