Services
Generating results is only part of an analysis. What the results mean, how much confidence to put in them, and whether they will hold in the next dataset is where most of the work goes.
Biomarker development and predictive modeling
Develop and evaluate gene-expression signatures and models using genomic and clinical data.
Deliverables: model code, performance estimates, and a clear account of validation and limitations.
Genomic analysis and interpretation
Run RNA-seq, immune-repertoire, and neoantigen workflows, then interpret the results in the context of the study.
Deliverables: quality-control summaries, analysis tables, figures, and documented methods.
Reproducible pipelines and reporting
Turn recurring analyses into documented, containerized workflows and reports that can be regenerated as data change.
Deliverables: code, environment definitions, and rerun instructions.
Independent implementation and review
Reimplement a signature, score, or analysis from its specification and compare it with the existing implementation.
Deliverables: documented agreement, discrepancies, and sensitivity to input changes.
Technical details
Signatures and models
- Signature development and scoring: signatures built from your data or taken from the literature, scored with ssGSEA, PLAGE and GSVA, with the gene-set and ontology work behind them.
- Outcome modeling: generalized linear and logistic models and Cox proportional hazards, univariable and multivariable, so a marker’s effect is reported next to the covariates competing with it.
- Multi-model comparison: elastic net, PLS, gradient-boosted trees, neural nets and GLMs fit under the same resampling. The model we recommend is chosen on predictive performance, stability across resamples, interpretability, and the use it is intended for, not on a single headline number.
- Prognostic and treatment-predictive signal: we distinguish associations with outcome from evidence that a biomarker identifies differences in treatment benefit, using appropriate treatment comparisons when the study design supports them. A marker can carry both kinds of information, and an association with outcome on its own does not establish differential benefit.
- Contrastive-loss elastic net: a specialized method that selects for treatment-predictive signal while suppressing prognostic effect size. It is useful where a randomized comparison exists to fit and judge it against; it does not by itself establish that a signature is treatment-predictive.
- Peptide–MHC binding models: affinity prediction feeding neoantigen and vaccine candidate selection.
Pipelines we run and interpret
- Vaccine and neoantigen candidate selection: LENS and RAFT, variant and expression evidence assembled per sample, candidates ranked with the filters stated rather than buried.
- RNA-seq quantification: through differential expression, signature scoring, and the comparisons the study was designed to make.
- TCR and BCR repertoire: clonality and diversity profiling, including from targeted and bulk expression data rather than dedicated deep sequencing.
Validation
- Held-out and external cohorts: candidate signatures are tested in data they were not fit on, and in an independent cohort when suitable independent data are available. Where no such cohort exists, we say so and report the resampling evidence instead.
- Sensitivity to the input: how scores respond to missing genes, batch structure, and the preprocessing choices upstream of them.
Engineering: containerized R and Python workflows, pinned third-party tool versions with recorded source URLs and checksums, per-step input and output manifests, and report generators that rebuild from the code rather than by hand.
Selected client work
Described at the level of what was built. Client data, results and findings stay with the client.
Immune profiling from existing data
Reading an immune repertoire normally calls for dedicated deep sequencing. We developed methods to estimate T-cell receptor diversity from targeted expression data the program was already generating, benchmarked the available reconstruction tools against each other, and established the quality-control conditions under which those estimates hold across sample types and platforms.
Read the case study →
Reproducible preclinical analysis
A preclinical program re-cuts the same in-vivo dataset as new experimental data arrives. We built a containerized workflow for patient-derived xenograft studies covering survival and tumor-growth modeling, expression quality control and signature scoring, with a report generator that emits a full report and a changes-only edition, every figure traceable to the script that drew it.
Read the case study →
Independent implementation and benchmarking
Reveal Genomics asked for an outside implementation of their breast-cancer prognostic assay. We reproduced the gene-expression signature and risk-score computation from the specification rather than from their code, measured how the scores respond when genes are missing, and benchmarked their shallow whole-genome sequencing analyses across paired plasma and tumor samples.
Read the case study →Analyses your team can inspect and rerun
We document the analysis methods, connect reported results to their source data and code, and provide the software environment and instructions needed to rerun the work. The project scope defines the checks and deliverables, including how discrepancies and limitations will be reported.
Scientific decisions, and responsibility for them, stay with Dante and with your team: thresholds, methods and what counts as adequate evidence are agreed and recorded rather than chosen silently.
- Figures and tables traceable to the data and the code that produced them.
- A containerized environment and instructions to rerun the analysis without us.
- Statistical choices stated and examined against the study design.
- A written record of what was checked, what was not, and what the checks cannot establish.
About
Dante Bortone, Ph.D. is a senior bioinformatics scientist at UNC Lineberger Comprehensive Cancer Center. His work combines genomic and clinical data, machine learning and survival analysis to study biomarkers for cancer immunotherapy. He also develops reproducible analyses, and is a co-author of the publications describing LENS and the use of LENS through RAFT for neoantigen analysis.
Dante trained in experimental neuroscience and earned his doctorate in neurobiology at UNC Chapel Hill. His research has included software for experimental control, imaging, stimulation and recording, alongside work he co-first-authored in Nature. He holds an issued patent and is a named inventor on a pending application.
That experience informs his consulting: understanding the experiment, choosing appropriate analyses, and delivering results that other scientists can examine and build on.
Selected publications
- Vensko SP II, Bortone D, Vincent BG. Predicting tumor antigens using the LENS workflow through RAFT. Methods Mol Biol. 2025;2932:319–342. PubMed
- Vensko SP, Olsen K, Bortone D, et al. LENS: Landscape of Effective Neoantigens Software. Bioinformatics. 2023;39(6):btad322. PubMed
- Bortone DS, Woodcock MG, Parker JS, Vincent BG. Improved T-cell receptor diversity estimates associate with survival and response to anti-PD-1 therapy. Cancer Immunol Res. 2021;9(1):103–112. PubMed
- Thorsson V, Gibbs DL, Brown SD, Wolf D, Bortone DS, et al. The immune landscape of cancer. Immunity. 2018;48(4):812–830.e14. PubMed
Projects
Selected software and product projects, showing Dante’s work in data modeling, controls, and practical product design.
SubTyper
An in-house Sabragen project
An interactive platform for organizing and sharing subtype and part-of relationships, with sources attached to individual claims. Ran 2017–2024; subtyper.com is currently offline, and a recorded demonstration is on the projects page.
More on SubTyper →Dawnward
Dante’s own venture, not a Sabragen engagement
A solar-tracking design that suspends panels from structures that already exist, instead of from a ground mount. Dante built the prototype, the control software and the embedded systems, and is the sole inventor on the issued patent.
More on Dawnward →DoDones
Dante’s own venture, not a Sabragen engagement
Physical task frames with sliding covers, supported by printable task cards and a web editor. The free printables and the editor are live; the shop is not open yet.
More on DoDones →