Skip to content
Isabela Lorente Kraetzer smiling in front of a grey wall.

Forensic genetics and bioinformatics

Isabela Lorente Kraetzer

Biomedical Sciences undergraduate at UNESP and CNPq research scholar. I build software that prepares genotype data for forensic analysis.

See the researchEmail me

Research profile

I'm a CNPq research scholar at GeMBio, the Molecular Genetics and Bioinformatics Laboratory at UNESP's Faculty of Medicine in Botucatu. I study genetic variation with forensic questions in mind: which markers could tell apart the people in a DNA mixture, what a genotype says about someone's appearance, and which populations a sample's ancestry points to.

Lab
GeMBio, Molecular Genetics and Bioinformatics Laboratory
Funding
CNPq undergraduate research scholarship
Degree
BSc Biomedical Sciences, UNESP, 2024-2027 (expected)
Based in
Botucatu, São Paulo, Brazil
  • Forensic genetics
  • Population genetics
  • Computational genomics
  • Bioinformatics
  • Genetic variation
  • Machine learning applied to genomic data

Research

KIRGenes studied
  1. KIR3DL3
  2. KIR3DP1
  3. KIR2DL4
  4. KIR3DL2

5,352individuals

79populations

Datasets

  • 1000 Genomes Project
  • HGDP
  • SABE

KIR region, simplified. The four framework genes I study are raised and labeled.

Sep 2026 - presentOngoing

Can KIR genes help untangle DNA mixtures?

Forensic samples often carry DNA from more than one person. I'm testing whether four highly variable KIR genes could work as extra markers for telling contributors apart. The data covers 5,352 people from 79 populations.

Analysis and validation are still in progress.

  • Undergraduate researcher, GeMBio, UNESP Botucatu
  • Funding: CNPq undergraduate research scholarship
  • I study hypervariable regions of the KIR gene complex as possible complementary markers for DNA mixture deconvolution.
  • I look at variation in KIR3DL3, KIR3DP1, KIR2DL4 and KIR3DL2 across the 1000 Genomes Project, HGDP and SABE datasets.
  • I process these highly polymorphic regions with bioinformatics tools, including kir-mapper.

Tools and data

  • kir-mapper
  • 1000 Genomes Project
  • HGDP
  • SABE
QCExample data, not from a real sample

Six example VCF records go through the same four checks the platform runs.

  1. VCF file
  2. Quality control
  3. HIrisPlex-S phenotypeSNPforID PLEX-34 ancestry
IDMarker IDRequired variantMissing genotypeREF/ALT alleles
rs12913832A>G 0/1Not run yetNot run yetNot run yetNot run yet
rs1805007C>T 0/0Not run yetNot run yetNot run yetNot run yet
.C>T 0/1Not run yetNot run yetNot run yetNot run yet
rs4988235G>A 1/1Not run yetNot run yetNot run yetNot run yet
rs12896399G>T ./.Not run yetNot run yetNot run yetNot run yet
rs16891982C>. 0/0Not run yetNot run yetNot run yetNot run yet

Press Run QC to check the records.

Jul 2025 - Sep 2026

A web app for forensic genotype data

The platform takes a VCF file, checks it and prepares it for two analyses: HIrisPlex-S phenotype prediction and SNPforID PLEX-34 ancestry inference. I built it, including the VCF processing and quality-control steps.

  • Undergraduate researcher, GeMBio, UNESP Botucatu
  • Funding: CNPq undergraduate research scholarship
  • The QC step checks marker identifiers, required variants, missing genotypes and REF/ALT alleles before any prediction runs.
  • It prepares genotypes for HIrisPlex-S phenotype prediction and for the SNPforID PLEX-34 panel for biogeographical ancestry.
  • I presented the results at CIC 2026, UNESP's undergraduate research conference.

Built with

  • Python
  • FastAPI
  • React
  • TypeScript
  • cyvcf2
  • pandas
Open the platform (opens in a new tab)See the CIC 2026 presentation
STRUCTURETrainingTest
Round 1
  1. Test
  2. Training
  3. Training
  4. Training
  5. Training
Round 2
  1. Training
  2. Test
  3. Training
  4. Training
  5. Training
Round 3
  1. Training
  2. Training
  3. Test
  4. Training
  5. Training
Round 4
  1. Training
  2. Training
  3. Training
  4. Test
  5. Training
Round 5
  1. Training
  2. Training
  3. Training
  4. Training
  5. Test

In five-fold cross-validation, each fold takes one turn as the test set.

Models compared
Decision TreeRandom ForestLogistic RegressionSVMXGBoost
Compared against
STRUCTURE
Scored with
F1-score, Balanced accuracy
Jun 2026

Can a classifier stand in for STRUCTURE?

I compared five machine-learning models as alternatives to STRUCTURE for ancestry inference with SNPforID PLEX-34 markers, scoring them with five-fold cross-validation, F1-score and balanced accuracy. Genetically similar populations were hard to tell apart, and admixed American (AMR) ancestry was hard to represent.

  • Poster project, Poster in Santos, Brazil
  • I tested Decision Tree, Random Forest, Logistic Regression, Support Vector Machine (SVM) and XGBoost models.
  • All models used the SNPforID PLEX-34 markers and were compared with five-fold cross-validation.
  • I presented this work as a poster in Santos in June 2026.

Methods

  • Decision Tree
  • Random Forest
  • Logistic Regression
  • SVM
  • XGBoost
  • 5-fold cross-validation
See the Santos poster

Where I've presented

Isabela at the XXXVIII CIC Unesp backdrop, wearing her congress badge.

Oral presentationSep 21-23, 2026

Development of an Application for Visualizing Phenotype and Ancestry Prediction Based on Genotypes

Authors: Kraetzer, I. L.; Ferreira, M. R.; Castelli, E. C.

XXXVIII UNESP Undergraduate Research Conference (CIC 2026)

Selected to advance to the second phase of the conference.

Isabela next to her e-poster at Biomedicina in Santos.
Isabela in front of the Biomedicina in Santos banner at the Santos Convention Center.

PosterJun 4-6, 2026

Evaluating the Performance of a Machine-Learning Model for Ancestry Prediction

Authors: Kraetzer, I. L.; Castelli, E. C.; Toloi, J. V. H.; Ferreira, M. R.

Biomedicina in Santos, Santos, Brazil

What I work with

Each well is a tool or method from my research. Select one to see where I used it.

Programming and data

Bioinformatics

Pipeline tools

Machine learning

Web development

Research practice

Toolbox plateRow

Select a well to read it.

From nursing technician to genomics

  1. Languages

    Portuguese and English

    Native and proficient

    Portuguese is my native language. My English is at a proficient level.

  2. 2016-2018

    Nursing technician

    COTUCA, University of Campinas (UNICAMP)

    Technical qualification in nursing at Colégio Técnico de Campinas.

  3. 2024

    BSc in Biomedical Sciences

    UNESP, Botucatu

    Started the degree. Graduation is expected in 2027.

  4. 2025

    Research at GeMBio

    Affiliated with UNESP's Faculty of Medicine, Botucatu

    Joined the Molecular Genetics and Bioinformatics Laboratory in July with a CNPq undergraduate research scholarship.

  5. 2025

    Accessibility and Inclusion

    UNIVESP

    80-hour course from the Virtual University of the State of São Paulo.

  6. 2025

    Human Reproduction Science

    COIL project

    International course with the University of Aveiro (Portugal), UNESP and UNIFIPA.

  7. 2026

    Facial Aesthetics and New Technologies

    UNESP

    8-hour course.

  8. 2027

    Graduation

    UNESP, Botucatu

    Expected end of the BSc in Biomedical Sciences.

Teaching and student leadership

Teaching

  • 2025-presentCurrent

    Volunteer instructor

    Cursinho Athena, UNESP Botucatu

    I teach in an outreach prep course for students applying to university.

  • 2024-presentCurrent

    Volunteer English instructor

    Let's Go Project, UNESP Botucatu

    I help children in public elementary schools learn English.

Representation and organizing

  • 2024-2026

    Student representative

    Biomedical Sciences Course Council (CONBIOMED), UNESP Botucatu

    I represent students in course-level discussions and take part in academic and administrative decisions.

  • 2026

    Director of registration

    Congresso de Biociências (CONGREBIO), UNESP Botucatu

    I run registration and help with the congress's organization, communication and day-to-day coordination.

Open to research opportunities

I'm looking for research internships, exchange programs and scholarships in genetics and bioinformatics. Email is the fastest way to reach me.

LinkedIn (opens in a new tab)Lattes CV (opens in a new tab)