For research teams

Research advisory and data science

Strategic research guidance and applied data science, delivered personally by Dr. Delva for academic teams, clinical groups and biotech organisations. Not templated, not delegated.

See sample deliverables Book a free 20-min call

What I offer

Two disciplines, one client relationship

Research advisor

Over a decade across oncology, molecular biology and neuroscience, from bench to strategy.

Research consulting & study design: experimental strategy, hypotheses and methods, before you commit resources.
Grant support & scientific writing: grant proposals (DFG, ERC, NIH) and manuscripts, grounded in what reviewers look for.
Academic mentoring: doctoral and postdoc coaching, fellowship applications and career transitions.

Data scientist

Applied statistics and machine learning, with data analytics training from Le Wagon, proven on real biomarker data.

Data analysis & visualisation: exploratory analysis, statistical testing and publication-quality figures in reproducible Python/R pipelines.
Machine learning & predictive modelling: classification and regression, feature engineering, validation, from raw data to an interpretable model.
Reporting & dashboards: BI dashboards, GA4/GTM reporting, SQL/BigQuery pipelines and data storytelling.

PythonSQLpandasscikit-learnRGraphPadStreamlit

Prices for research teams →

Engagement tiers

Three levels of depth

The same structure applies to a research advisory sprint or a full modelling pipeline. Each project is quoted after a free call.

★

Starter

Exploratory report or literature/methods review
3 publication-quality figures
Reproducible notebook or annotated draft
Written findings and next steps

★★

Standard

Everything in Starter
Multivariable modelling or structural revision
Subgroup and interaction analyses
5+ figures or tables; 1 round of revisions

★★★

Advanced

Full ML pipeline or full publication package
Model comparison and validation, or journal targeting
6+ figures with biological interpretation
2 rounds of revisions

Sample deliverables

What a client receives

★ Starter · Case study 01

Exploratory Data Analysis Report

Sleep quality and stress biomarker dataset · N = 180 participants

6.8 hrsMean sleep duration

r = −0.43Sleep–stress correlation

38%Below recommended sleep

5 daysTurnaround

A rigorous statistical summary of how sleep duration, stress and intervention group relate, with publication-quality figures and an interpretation grounded in current sleep science. Python (pandas, NumPy, SciPy, Matplotlib); Pearson correlation, t-test, Cohen's d.

Measure	Value	p	Interpretation
Mean sleep duration	6.8 hrs	—	Below the recommended 7–9 hrs for adults
Mean stress score	5.2 / 10	—	Moderate stress across the sample
Sleep–stress correlation	r = −0.43	< 0.001	Significant negative correlation
Intervention vs. control sleep	+0.8 hrs	< 0.05	Modestly higher sleep in the intervention group
Group effect (Cohen's d)	0.38	—	Small to moderate effect
Under 6 hrs sleep per night	38%	—	Below the minimum recommended threshold

Recommended next steps

Follow-up data at 3 and 6 months to test durability
Add validated sleep quality measures (PSQI)
Control for age, gender and baseline stress in multivariable models
Subgroup analyses to find the strongest responders; expand to N ≥ 400

Delivered: written report (PDF + Word), reproducible Jupyter notebook, 3 figures at 300 DPI.

★★★ Advanced · Case study 02

ML Predictive Analysis Report

Sleep disorder risk classification · N = 300 participants · 7-feature biomarker dataset

0.847AUC-ROC (Random Forest)

78%Held-out accuracy

3Models compared

A full machine learning pipeline predicting sleep disorder risk (high vs. low) from sleep, stress, age, cortisol, BDNF, CRP and IL-6. Validated with 5-fold cross-validation and a held-out test set; feature importance and partial dependence connect the model back to the biology.

Model	AUC-ROC	Accuracy	CV score
Logistic regression (baseline)	0.791	72.3%	0.784 ± 0.04
Gradient boosting	0.834	76.7%	0.826 ± 0.03
Random forest (selected)	0.847	78.0%	0.839 ± 0.03

Key findings

Sleep duration and stress score carry 50% of the model's discriminative power; both are modifiable targets.
Below 6 hours of sleep, predicted risk rises sharply; the 6–7 hour window is the highest-leverage zone.
Cortisol and BDNF add independent predictive value beyond sleep and stress.
Sleep duration + stress + cortisol capture 68% of the full model's predictive power, a lightweight screening option.

Delivered: full report (Word + PDF), Jupyter notebook, 6 figures, model evaluation spreadsheet and CSVs, 2 rounds of revisions.

Have data or a study that needs a second scientist?

Projects are scoped to your data, timeline and budget.

Book a free call See prices