Your next R&D milestone starts with better evidence.

From disease research to diagnostic development, BioBuild provides the bioinformatics expertise, tailored analyses, and reproducible results needed to turn complex biological data into actionable insights.

biobuild · workspace

  • Pull cohort from GEO + internal storeGSE-neuro · n=42 · checksum verified
  • Normalize & run differential expressionDESeq2 v1.42 · params logged
  • Flag candidate biomarkers & plotFDR < 0.05 · figure_3.svg
Provenance sealed sha256:9f3c…a71b

BioBuild / Client work

Working with teams across life sciences

BioBuild turns complex biological data into results that guide the next R&D decision.

Using assay data and machine learning to guide diagnostic development

Astek needed to understand what its plate-reader data could reveal about rapid urinary tract infection testing and which approaches deserved further development.

Explore the story

01 / Astek Diagnostics

The challenge

Astek needed to understand what its plate-reader data could reveal about rapid urinary tract infection testing and which approaches deserved further development.

How we helped

BioBuild analyzed the experimental data and developed machine learning models to investigate patterns in the measurements. We compared findings against available reference measurements and assessed where further evidence was needed.

What the team gained

The results helped Astek prioritize analytical directions and design follow-up experiments to test the remaining questions.

Using single-cell data to guide neuroimmune research

Researchers wanted to understand how chronic stress affects microglial populations and gene activity, and use those findings to focus their next experiments.

Explore the story

02 / Johns Hopkins School of Medicine

The challenge

Researchers wanted to understand how chronic stress affects microglial populations and gene activity, and use those findings to focus their next experiments.

How we helped

BioBuild analyzed single-cell data, annotated cell populations and compared gene-expression patterns across conditions. We tested biological hypotheses and interpreted the findings alongside the team’s research questions.

What the collaboration established

The results helped the researchers prioritize hypotheses and design follow-up studies.

Abstract illustration of cells

Connecting preclinical data to development decisions

An oncology team needed to evaluate potential targets and indications using evidence from efficacy studies, immune-response experiments and molecular profiling.

Explore the story

03 / Emerging Oncology Company

The challenge

An oncology team needed to evaluate potential targets and indications using evidence from efficacy studies, immune-response experiments and molecular profiling.

How we helped

BioBuild analyzed the datasets, tested the team’s hypotheses and brought the results together to assess candidate development directions and identify unanswered questions.

What the team gained

The findings helped the team prioritize hypotheses and design follow-up studies to support its next development decision.

Who we work with

Bring the data your instruments produce.

Assay developers, clinical-stage biotechs, discovery teams, CROs and core facilities. Pick the one closest to you. The lists are examples, not limits.

Tell us what you have

Typical asks

Detection limits and repeatability for a regulatory file, concordance with a reference platform, lot-to-lot drift.

Data we typically receive

  • Multiplex immunoassay panelsLuminexMSD
  • Affinity proteomicsOlink NPXSomaScan ADAT
  • Targeted sequencing panelsFASTQBAMVCF
  • qPCR and ddPCRCt tablesRDML
  • Detection limit and repeatability studiesCLSI EP17EP05
  • Cross-platform concordanceBland-AltmanLin’s CCC

Not listed? Most of what we receive isn’t. Send the format and the question.

Biological insights

Uncovering the biology behind the numbers.

We start with the biological question and shape the analysis around the evidence needed to investigate it. These examples show how we use tissue context, findings across studies and cell identity to understand biological signals in published data.

biobuild · analysis explorerRecorded example
Research question

Where do injury and macrophage-associated signals appear in kidney tissue, and how do their spatial patterns differ between the sampled conditions?

GSE269622 · 10x Visium · Sham, AKI 4 h and AKI 12 h

Study setup · acute kidney injury
ShamControl section
4 hoursEarly injury
12 hoursAcute response
Visium tissue sections → spatial gene-expression data

Original schematic of the analysed subset: one section per condition. Kidney drawings are condition schematics, not histology. Based on Xuanyuan et al. (2025); the published study includes additional timepoints and Xenium profiling.

Example requests summarise the reviewed studies. This walkthrough displays recorded results; it does not run a new analysis.

Original case-report figure

What you get when you hand us a dataset.

Each case starts with a question a research team would bring us. We run the analysis end to end to uncover deeper insights and reliable findings. See how those results can shape your next study and guide your next R&D milestone.

Research stories · Transcriptomics01 / 04

RNA-seq:
Which findings
hold across pipelines?

Goal

Different processing choices can change which genes appear significant.

Solution

Compared five workflows on the same hypoxia dataset.

Result / Transcriptomics

489 / 489 shared calls agree in direction.

Runs A and D

All 489 genes called in both runs A and D agreed in direction. The comparison also identified genes with workflow-dependent support.

01 / Figure

Selected hypoxia-response genes

Ten selected hypoxia-response genes across five workflows. Eight show consistently positive, significant effects; CA9 and SLC2A1 show run-dependent support. Exact log2 fold changes and adjusted P values are in the CSV.

Ten selected response genes across five workflows in LNCaP cells, with two samples per condition. Eight show consistently positive, significant effects; CA9 and SLC2A1 show run-dependent support. Values show log2 fold changes. DESeq2 estimates use apeglm shrinkage.

02 / Key numbers

What the analysis covers.

489 / 489Shared calls agreeruns A and D
5Workflowscompared
10Genesselected
Methods and valuesShow details
Log2 fold change per workflow; adjusted P values are in the CSV.
GeneABCD1D
NDRG12.822.962.962.972.91
STC15.664.455.786.235.76
PLOD25.616.06.015.925.82
PFKFB44.063.073.083.243.06
LDHA2.062.091.821.841.83
BNIP31.972.012.072.12.13
EGLN32.352.332.592.372.7
VEGFA1.651.711.781.651.7
CA90.021.842.850.132.64
SLC2A12.62.191.460.141.58

The same four empty-vector LNCaP samples were analyzed five ways: A (one SRA run per sample), B (deposited RPKM with limma-trend), C (all four runs with GENCODE v38), D1 (A’s reads with updated processing), and D (all four runs with Ensembl r111).

Gene-call threshold: adjusted P < 0.05 and |log2FC| > 1, using shrunk effects for DESeq2 runs. A and D share 489 calls; all 489 agree in direction. The 165 A-only and 535 D-only calls show that call sets also differ.

Depth, annotation, software and processing differ across comparisons. More calls do not establish greater accuracy. The five workflows analyze the same samples, not independent biological replicates. CA9 A and SLC2A1 D1 are not significant; CA9 D1 has no adjusted P after independent filtering.

GEO GSE106305BioBuild hypoxia analysis report, selected-gene table.
Explore the original source ↗
Research stories · Spatial biology02 / 04

Spatial biology:
Where are the
expression patterns?

Goal

A multi-organ cancer tissue array contains expression patterns that need spatial context.

Solution

Analyzed a Visium HD lung, breast and colon cancer array, linking expression groups to tissue locations and marker evidence.

Result / Spatial biology

573,965 retained 8 µm bins.

Eight expression groups

Organized 573,965 retained 8 µm bins into eight expression groups, with spatial maps and marker summaries.

01 / Figure

Expression groups in tissue

Original spatial map of 200,000 sampled 8 micrometer bins from a multi-organ cancer tissue array, colored by eight marker-informed expression groups. These are not validated cell types.

Map displays 200,000 sampled bins. 100,000 bins directly clustered; labels projected to 573,965 retained bins. Group identities are marker-based interpretations, not validated cell types. Credit: 10x Genomics.

02 / Key numbers

What the analysis covers.

573,965Retained bins8 µm
8Expression groupsmarker-informed
200,000Bins plottedsampled
Methods and valuesShow details
Retained 8 µm bins per expression group; label-confidence notes are in the CSV.
GroupRetained binsReported %
0134,91823.5
1107,66218.8
295,70216.7
390,85515.8
449,7738.7
544,5347.8
626,6354.6
723,8864.2

Of 1,290,617 in-tissue bins, 573,965 have at least 100 total UMI counts and were retained. A random subsample of 100,000 bins (seed 42) was processed with PCA, UMAP and Leiden clustering (resolution 0.5).

Labels were projected to all retained bins using 15-nearest-neighbour classification in PCA space. Projection quality was not reported. The eight groups may reflect tissue of origin as well as cell-associated expression.

Candidate labels come from marker overlap and manual refinement, without pathologist review or orthogonal validation. Groups 2 and 5 have low-confidence interpretations in the review. Bins are not single cells.

10x Genomics Visium HD cancer tissue array (CC BY 4.0)BioBuild P09 analysis report, spatial-map figure.
Explore the original source ↗
Research stories · Biomarker assessment03 / 04

Kidney cancer:
Which genes warrant further investigation?

Goal

Determine whether candidate genes differ between tumour and normal tissue, and whether CASR expression is associated with survival.

Solution

Compared six genes in 72 matched tissue pairs and analyzed survival in 529 TCGA-KIRC cases.

Result / Biomarker assessment

72 pairs, six genes compared.

Paired expression comparison

All six genes differed between tumour and normal tissue at P < 0.001. Higher CASR expression was associated with lower mortality hazard: HR 0.715, 95% CI 0.53–0.96.

01 / Figure

CASR expression and overall survival

Higher versus lower CASR expression has an estimated mortality hazard ratio of 0.715, with 95 percent confidence interval 0.53 to 0.96, in 529 TCGA-KIRC cases. This is an association, not a treatment effect.

CASR high versus low expression, median split: 529 cases, 173 deaths. Cox HR 0.715 (95% CI 0.53–0.96). Survival uses diagnosis and follow-up records. An observational association, not a validated prognostic test.

02 / Key numbers

What the analysis covers.

72 pairsMatched tissuetumour and normal
529CasesTCGA-KIRC survival
0.715Hazard ratio95% CI 0.53–0.96
Methods and valuesShow details
CASR high versus low median split; overall survival, TCGA-KIRC.
MeasureValue
Hazard ratio0.715
95% CI0.53–0.96
High-CASR group265
Low-CASR group264
Cases529
Deaths173
Log-rank P0.0276
Cox Wald P0.028

Expression comparison used 72 matched tumour/normal cases, a two-sided paired t-test, and log2(TPM + 1). All six candidate genes had P < 0.001.

The survival analysis used one primary-tumour sample per case: 529 cases with usable overall-survival data, 173 deaths, and a median CASR split of 265 high versus 264 low. Time was days to death or the latest follow-up day from diagnosis and GDC follow-up records.

The reported Cox HR is 0.715 (95% CI 0.53–0.96); log-rank P = 0.0276. A diagnosis-only first pass omitted many censored cases and gave a different result; the reported result includes follow-up records.

Survival was examined for CASR only. This unadjusted association does not establish causality, clinical validity or a validated prognostic test. No reference-paper figure is reproduced.

Source: BioBuild TCGA-KIRC statistical analysis report, pp. 1–3; reference comparison: Cui et al. (2020), doi:10.3892/ol.2020.11703. Credit: TCGA Research Network.

TCGA-KIRCBioBuild statistical analysis report, CASR survival.
Explore the original source ↗
Research stories · Single-cell analysis04 / 04

Lung single-cell:
Which markers
distinguish cell populations?

Goal

Identify genes that distinguish cell populations in lung tumour and normal tissue.

Solution

Analyzed 88,100 cells from 22 samples across eight published cell-type groups.

Result / Single-cell analysis

24 / 24 canonical markers checked.

Recovered in expected groups

Produced 16,803 candidate gene–cell-type marker entries and recovered all 24 canonical markers checked in their expected groups.

01 / Figure

Canonical-marker consistency check

All 24 selected canonical markers were recovered in their expected published cell-type groups. The chart shows the range of AUC values across three checked markers per group; these ranges are not confidence intervals.

Marker re-run: 88,100 cells from 22 samples, using eight cell-type labels supplied by the original authors. AUC ranges summarize three checked canonical markers per group; they are not confidence intervals or an independent annotation benchmark.

02 / Key numbers

What the analysis covers.

24 / 24Markers recoveredcanonical
88,100Cellsanalyzed
22Samplestumour and normal
Methods and valuesShow details
AUC range across three canonical markers per group; not a confidence interval.
Published cell typeCanonical genesAUC minAUC max
T lymphocytesCD3D / CD3E / CD20.830.91
Myeloid cellsCD68 / LYZ / CD140.780.97
Epithelial cellsEPCAM / KRT18 / NAPSA0.880.92
NK cellsNKG7 / GNLY / KLRD10.890.98
B lymphocytesMS4A1 / CD79A / CD79B0.720.94
FibroblastsCOL1A1 / DCN / LUM0.910.98
MAST cellsTPSAB1 / KIT / CPA30.761.00
Endothelial cellsPECAM1 / VWF / CLDN50.840.95

The public GEO count matrix and original authors’ cell annotations were used for 11 tumour and 11 normal lung samples. The featured marker re-run retained 88,100 cells. Published cell-type labels were inherited, not newly discovered.

Candidate markers were tested one published cell type versus the other cells, using Wilcoxon rank-sum testing on log-normalized data, with log2FC > 0.25, min.pct 0.1 and raw P < 0.01. The result contains 16,803 gene–cell-type entries, not 16,803 unique genes.

All 24 selected canonical markers were present in the expected group lists. AUC summarizes separation for the checked markers, not a general classifier accuracy score. The chart’s per-group range is not a confidence interval.

This cell-level marker comparison does not establish patient-level differential abundance or independent biological validation. The original first-run figures used 88,103 cells; that separate run is not the basis of the marker-run counts shown here.

GEO GSE131907 — Kim et al. (2020)BioBuild lung single-cell analysis report, pp. 1–2 and 6–7. Source study: Nature Communications 11:2285 (CC BY 4.0).
Explore the original source ↗

The four case studies use public datasets to show our methods. They are not client results, clinical products or experimental validation.

Technical details

BioBuild plans the analysis, runs real code on your data and checks its own numbers. You approve the plan first, and every result links back to the step that produced it.

Human scientist

Sets the question, approves the plan, and signs off on every result.

Your data

Sequencing, imaging and tabular files you upload

Public databases

NCBI, Ensembl, UniProt, gnomAD and 5 more

Literature and web

Search results the agent reads and cites

BioBuild agent Running
  1. PlanDrafts the analysis plan and waits for your approval before any code runs.
  2. Skills36 domain skill packs: single-cell, proteomics, CRISPR, metagenomics and 32 more.
  3. ExecuteRuns the code in an isolated cloud sandbox with your files mounted.
  4. VerifyRecords every reported number with its source file and the script that computed it.
Report

Findings with figures, tables and citations

Figures and data

Every file a run produced, ready to download

Capsule

A re-runnable record of each step, its inputs and its environment

Who we help

CRO teams

Add specialist analysis capacity to your next study.

Scope a defined analytical workstream with BioBuild, from data review through figures, interpretation, and a documented handoff. Align the outputs with your study question and client reporting needs.

Scope your study analysis

Biotech teams

Get the readout your next program decision needs.

Bring the data you already have and the decision it needs to inform. We help assess the findings, prioritize follow-up, and prepare an analysis package your team can examine and build on.

Discuss your next readout

What does your data need to answer?

For CRO teams and lean biotech: bring your research question, existing data, and the decision behind your next milestone.

Discuss your analysis

Talk through your question, inputs, and desired outputs.