Marcelo Mollinari
  • Quantitative genetics
  • Plant breeding
  • Scientific software

Marcelo Mollinari, PhD

Quantitative Geneticist / Consultant at AbacusBio

My work spans statistical methodology, scientific software, and applications of quantitative genetics to complex breeding populations, bringing together genetic theory, computational methods, and practical statistical analysis. It includes MAPpoly and MAPpoly2, open-source software for linkage mapping and haplotype inference in diploid and polyploid populations.

Previously Research Assistant Professor, North Carolina State University.

Selected work Get in touch
Portrait of Marcelo Mollinari

  • Statistical methods

    Hidden Markov, random-effect, and mixed models for mapping, QTL analysis, genetic evaluation, and prediction.

  • Working with breeders

    From research programs in polyploid crops to commercial breeding in horticultural crops, forest trees, and livestock.

  • Usable tools

    Open R and C++ packages, and reproducible workflows that make analyses accessible to breeders.

Selected scientific work

Public research and software projects, from statistical models to the tools that put them to use.

Genetic mapping · Polyploids

Multi-family linkage maps in polyploids

Creator and lead developer of MAPpoly and MAPpoly2.

Problem
Breeding programs often work with several related crosses. Linkage maps built one family at a time contain partly different sets of markers and use only the information within each family.
Approach
MAPpoly2 extends the single-family models in MAPpoly to interconnected diploid and autopolyploid families. A hidden Markov model builds a map for each parent. These are combined into a joint map, which is then recomputed to place the remaining markers. Ploidy levels 2, 4, and 6 can be combined in any configuration, and the computationally intensive steps are written in C++.
Contribution
Integrated linkage maps that combine information across interconnected families, implemented as open-source R software. The example output below is a consensus map for a multiparental potato population.

MAPpoly2 on GitHub MAPpoly method (G3, 2019)

Integrating maps from several populations Three populations each have markers mapped separately, at partly different positions. An arrow leads to a single consensus map that contains the markers from all three populations. Pop 1 Pop 2 Pop 3 Consensus
Maps from separate populations combined into one consensus map (simplified).
Original figure: how MAPpoly2 builds the map
Original figure: markers mapped separately in three populations are linked through a hidden Markov model into a single consensus map, with the full caption below the diagram.
Original figure with its full caption.
Example output: a potato consensus map
MAPpoly2 output for a multiparental potato population: a population summary table, per-family and consensus maps for 12 chromosomes, and the final consensus linkage map.

Recent methods development

Maternal maps from open-pollinated families

Problem
In open-pollinated families the mother is known and genotyped, but the fathers are not observed. Persistent paternal haplotype-sharing tracts, as arise when some pollen parents are related to the mother, can otherwise be absorbed into the maternal recombination map.
Approach
A source-aware hidden Markov model pairs the transmitted maternal homolog with a paternal source state, using the offspring genotypes and the genotyped mother. Paternal haplotype-sharing tracts are modelled explicitly. These states describe shared tracts statistically; they do not identify fathers. Intervals without linkage are reported as gaps rather than as large distances.
Contribution
HalfSibMap, an R package that builds maternal linkage maps and reconstructs each offspring's maternal inheritance in diploid open-pollinated families.
Role
Creator and lead developer, with co-authors Kim McKeever, Molly Irvin, and Fikret Isik.
Status
Available from GitHub; a methodological manuscript is in preparation.
Reconstructing maternal inheritance in an open-pollinated family The mother's two homologs are drawn as two bars; the fathers are not observed. Below, ten offspring are each drawn as a bar showing which maternal homolog they inherited along the linkage group, changing from one homolog to the other at each crossover. A maternal linkage map with marker positions runs along the bottom. homolog 1 homolog 2 Mother Fathers not observed Offspring Maternal map
Reconstructed maternal inheritance: which of the mother's two homologs each offspring carries along a linkage group (simplified).
Example output: one linkage group
HalfSibMap output showing a recombination heatmap, reconstructed maternal inheritance, and paternal-process summaries along a linkage group.
Example HalfSibMap output showing linkage mapping, reconstructed maternal inheritance, and paternal-process diagnostics.

View full-size figure (PDF)

Genomics · Sweetpotato

Inheritance and genome architecture of hexaploid sweetpotato

Problem
Hexaploid sweetpotato has six sets of chromosomes (2n = 6x = 90), making inheritance analysis more complex. Because of this, genetic studies in sweetpotato lagged well behind those in major diploid crops.
Approach
Ultra-dense multilocus mapping in a full-sib population from the cross Beauregard × Tanzania, using MAPpoly and more than 30,000 informative SNPs. Genotype probabilities were computed across the genome for every individual, and the complete hexaploid haplotypes of the offspring were inferred.
Contribution
An ultra-dense integrated genetic map with 96.5% collinearity with the diploid relative Ipomoea trifida. Inheritance was hexasomic in all linkage groups, apart from low levels of preferential pairing in linkage group 2; 73.3% of meiotic configurations were resolved as bivalents and 15.7% showed multivalent signatures. Related work includes a chapter on genetic maps in sweetpotato and, as a co-author, a phased chromosome-level genome assembly.

Inheritance (G3, 2020) Genome assembly (Nature Plants, 2025) Genetic maps in sweetpotato (2025)

Haplotype reconstruction and meiotic pairing in one hexaploid offspring Each parent, Beauregard and Tanzania, has six homologs, drawn as six coloured bars. The offspring has six homologs, three inherited from each parent. One is a whole parental homolog; the other five are built from segments of two or three homologs of the same parent, joined at crossovers. Each crossover is also marked on the parent, between the two homologs that exchanged segments. Two further views show what the crossovers say about pairing at meiosis. In Beauregard each crossover joins a different pair of homologs and no homolog has two partners, which is consistent with three bivalents. In Tanzania the homologs paired on the left part of the chromosome are not the ones paired on the right part: four homologs exchanged with two different partners. That change of partners joins all six homologs into one hexavalent, drawn as a star with six arms, each arm a pair of homologs, and the change of partners at its centre. A fourth view picks out, in the bivalents and in the star, the three chromosomes each meiosis sent to its gamete: each follows one homolog and passes to its partner at a crossover. In the last view the two gametes come together as the six homologs of the offspring. Beauregard Tanzania Offspring
One offspring of the Beauregard × Tanzania cross, linkage group 1. It carries three homologs from each hexaploid parent, most of them rebuilt from segments of that parent's six homologs. The crossovers also show how the homologs paired at meiosis: consistent with three bivalents in Beauregard, and with one hexavalent in Tanzania, where homologs changed partners along the chromosome (simplified).
Full example: homolog probabilities for this offspring
Original figure: probability profiles along about 300 centimorgans for the six Beauregard homologs, a to f, and the six Tanzania homologs, g to l, with arrows at the recombination points; beside them, the resulting gametes and the six reconstructed homologs of individual BT05.320.
Probability profiles of the 12 parental homologs for offspring BT05.320, linkage group 1, with the resulting gametes and the reconstructed haplotype. Part of Figure 3 in the 2020 G3 paper.

Visualization

Exploring polyploid genetic analyses in one place

Conceptual lead of VIEWpoly.

Problem
Results from linkage, QTL, and genome analyses in polyploids often sit in separate tools.
Approach
VIEWpoly is an interactive application that brings them together, so users can explore candidate genes in a genome browser. It reads output from MAPpoly, polymapR, diaQTL, QTLpoly, and polyqtlR.
Contribution
One place to explore linkage, QTL, and genomic results from five polyploid analysis packages, published in the Journal of Open Source Software (2022) and distributed on CRAN.

VIEWpoly paper (JOSS, 2022) Video tutorial

Interactive explorations in quantitative genetics

Quantitative genetics brings together statistical inference, genetic theory, and computation to understand biological variation and inform breeding decisions. These explorations take three ideas from that much larger field and make them tangible with simulated data.

Mixed models · Multi-environment trials

How do environmental and field effects influence estimates of genetic merit?

What is measured on a plot reflects more than genetic merit. The environment of the trial, the plot’s position in the field, and the way each genotype responds to that environment all contribute. Growing the same genotypes in several environments makes it possible to tell these sources apart. Here a mixed model splits three simulated trials into their parts.

Relationship matrices · Pedigree and markers

What can genomic information reveal about expected genetic relationships?

A pedigree describes the expected genetic relationship between individuals, while genomic markers provide an estimate of their genetic relationship based on observed marker information. The two are related but not identical: relatives do not share exactly the expected fraction of their genomes, and markers measure that fraction with some error. Comparing them shows this variation and, occasionally, pairs that do not fit the pedigree at all.

Reaction norms · Selection across environments

Is the best genotype the same in every environment?

Genotypes can differ in how they respond to the environment. When they do, the differences between them change from one environment to the next, sometimes only in size and sometimes in order. Each line below is one genotype’s genetic merit along an environmental gradient. Move the target to see which five would be selected there.

Scientific software

Open-source R packages I develop, maintain, and contribute to.

Selected publications

  1. Crowley J, Bjelland D, Hely F, Valente T, Mollinari M. (2026). Agricultural genomics data: assessing the opportunity for data standards. Canadian Standards Association, Toronto, ON.
  2. Wu S, Sun H, Zhao X, Hamilton JP, Mollinari M, Gesteira GDS, Kitavi M, Yan M, Wang H, Yang J, Yencho GC, Buell CR, Fei Z. (2025). Phased chromosome-level assembly provides insight into the genome architecture of hexaploid sweetpotato. Nature Plants, 11, 1951–1959. doi:10.1038/s41477-025-02079-6
  3. Gesteira GS, Pereira GS, Zeng ZB, Mollinari M. (2025). Genetic maps in sweetpotato. In The Sweetpotato Genome (pp. 45–68). Springer. doi:10.1007/978-3-031-65003-1_5
  4. Taniguti CH, Gesteira GS, Lau J, Pereira GS, Zeng ZB, Byrne D, Riera-Lizarazu O, Mollinari M. (2022). VIEWpoly: a visualization tool to integrate and explore results of polyploid genetic analysis. Journal of Open Source Software, 7(74), 4242. doi:10.21105/joss.04242
  5. Pereira GS, Gemenet DC, Mollinari M, Olukolu BA, Wood JC, Diaz F, Mosquera V, Gruneberg WJ, Khan A, Buell CR, Yencho GC, Zeng ZB. (2020). Multiple QTL mapping in autopolyploids: a random-effect model approach with application in a hexaploid sweetpotato full-sib population. Genetics, 215(3), 579–595. doi:10.1534/genetics.120.303080
  6. Mollinari M, Olukolu BA, Pereira GS, Khan A, Gemenet D, Yencho GC, Zeng ZB. (2020). Unraveling the hexaploid sweetpotato inheritance using ultra-dense multilocus mapping. G3: Genes, Genomes, Genetics, 10(1), 281–292. doi:10.1534/g3.119.400620
  7. Mollinari M, Garcia AAF. (2019). Linkage analysis and haplotype phasing in experimental autopolyploid populations with high ploidy level using hidden Markov models. G3: Genes, Genomes, Genetics, 9(10), 3297–3314. doi:10.1534/g3.119.400378

Full publication list on Google Scholar

About

I am a quantitative geneticist trained in plant genetics and breeding. My doctoral work at the University of São Paulo developed models for building genetic maps in autopolyploids, with applications in sugarcane. Since then I have focused on algorithms for genomic analysis in complex family structures, especially outcrossing diploid and polyploid crops.

At North Carolina State University I led research on statistical methods and computational algorithms for diploid and polyploid populations, created MAPpoly and MAPpoly2, and worked with research groups and breeding programs in sweetpotato, potato, blackberry, forage crops, and other species, combining statistical models with R and C++ software.

Since October 2025 I have worked at AbacusBio on quantitative genetics and biometrics for commercial breeding applications across horticultural crops, forest trees, and livestock.

Areas of focus

  • Statistical & quantitative genetics
  • Genomic prediction & mixed models
  • Linkage analysis & haplotype phasing
  • QTL mapping
  • Polyploid genetics
  • R & C++ package development

Research leadership and service

  • At NC State, project director on some externally funded research projects and co-investigator on others, with support from USDA-NIFA, USDA-SCRI, CGIAR, and other organizations. Across these projects I coordinated methodological development, software implementation, and scientific collaborations.
  • Graduate student advisor and committee member at NC State, and mentor to graduate students and researchers.
  • Peer reviewer for Genetics, G3, Nature Communications, Bioinformatics, and other scientific journals.
  • Member, National Association for Plant Breeding (NAPB).

Experience

  1. Quantitative Geneticist / Consultant

    AbacusBio · October 2025–present

  2. Research Assistant Professor

    Bioinformatics Research Center & Department of Horticultural Science, North Carolina State University · 2021–October 2025

    Earlier at NC State: Senior Research Assistant (2018–2021) and Postdoctoral Fellow (2016–2017)

  3. Postdoctoral Fellow

    Department of Genetics, ESALQ, University of São Paulo · 2012–2015

    Visiting Scholar, Department of Statistics, Purdue University (2014–2015)

  4. Research Intern, eScience Group

    Microsoft Research · 2010

Education

  • PhD, Plant Genetics and Breeding, ESALQ, University of São Paulo (2012)
  • MSc, Plant Genetics and Breeding, ESALQ, University of São Paulo (2008)
  • BSc, Agronomic Engineering, ESALQ, University of São Paulo (2005)

Talks, workshops & resources

Slides from talks, workshops, and lectures (PDF).

Scripts and guides

Contact

I welcome conversations about quantitative genetics, plant breeding, and breeding analytics, whether you work in a breeding program or company, are a researcher interested in collaborating, or use one of my software packages.

Email
marcelo@marcelomollinari.com
Google Scholar
Scholar profile

For software questions or bug reports, the issue tracker in each package's repository is usually the quickest route.