CacaoIQ™by Scilytics
CacaoIQ: from the field trial to the candidate gene.
Reads your phenotype and genotype files, works out how the trial was laid out, fits the model that design supports and takes the result down to the genes under each peak. Runs offline. It is set up for cacao clonal trials, full-sib families and the Matina 1-6 v1.1 reference genome.
A trial takes years to score. Analyzing it shouldn’t take weeks.
The usual way
Several tools and scripts, with the data passed between them.
- Reshape the phenotype file in a spreadsheet for each analysis
- Fit mixed models in R, with a script for each trial design
- Build linkage maps in a separate program, with its own file format
- Look up the genes under each peak on annotation websites
- Redo the steps by hand when the data change
With CacaoIQ
One desktop app, with no R, Python or Java to install.
- Load the phenotype and genotype files
- Confirm the columns and the proposed model
- BLUPs, GWAS, QTL mapping and candidate genes in one run
- A README with the settings of every run
- The same steps for the next trial
What CacaoIQ works out for you
CacaoIQ shows you what it worked out, and asks you to confirm the parts that depend on your trial.
Files in
Phenotypes as CSV, TXT or Excel (you pick the sheet). Genotypes as VCF, HapMap or an allele-dose table.
CacaoIQ finds the header row, suggests which columns are identifiers, traits and design factors, and recognizes local missing-value codes. You confirm the mapping.
Flags duplicates, outliers and impossible values, and suggests a transform when a trait is skewed.
Public data: Bekele et al. 2022 The model
Recognizes field rows and columns, incomplete blocks, repeated measures and factorial crosses from the columns in your file.
Proposes the model that design supports: spatial or block-corrected BLUPs, GCA and SCA for factorial crosses, and REML by default. Bayesian MCMC covers the traits REML doesn’t, such as presence/absence, counts and repeated measurements per tree.
Public data: Osorio-Guarín et al. 2020 GWAS and QTL mapping
Corrects GWAS for kinship, applies a Bonferroni threshold, and uses a genome-wide permutation threshold for QTL in full-sib families.
In a full-sib family with both parents genotyped, QTL mapping gives the QTL a LOD-support interval from the family’s own linkage map.
Candidate genes
Lists the candidate genes under each significant peak, quoted from the bundled UniProt and InterPro annotation. With a linkage map the search window is the QTL’s LOD-support interval. With GWAS alone it is 250 kb either side of each significant region, and the summary lists the genes nearest the peak, with every gene in the table.
Public data: Osorio-Guarín et al. 2020 Results out
Interactive charts, saved as PNG, and sortable tables, exported as CSV.
A plain-text README for each run: what was run, the settings and the key results, all of it in a results folder you choose.
A methods assistant that stays on your computer
Ask how a method works or what a line on a chart means, and the assistant answers from the CacaoIQ user manual, citing the section it used. When you load a file, it suggests a role for each column, and you confirm them. It runs on your computer like the rest of CacaoIQ, so your questions and your data stay with you.
Checked against published studies
A published cacao GWAS, rerun.
On the public data of Osorio-Guarín et al. (2020), all 10 markers the paper reports on the Matina 1-6 genome came out among CacaoIQ’s 12 strongest for their trait. All validation results
Trial statistics checked against published values.
The six trait correlations published by Bekele et al. (2022) are reproduced to three decimals. On a classic incomplete-block trial (John & Williams 1995 oat α-lattice) the genetic variance matches the reference value distributed with the data (0.14290 against 0.142902). For Baek et al. (2026), applying the paper’s own repeatability formula to the same data CacaoIQ reads gives the paper’s four values exactly. The paper reports the reliability of a four-harvest mean. CacaoIQ reports the standard single-measurement repeatability.
What is not checked yet.
Genomic heritability and GCA/SCA are not yet checked like-for-like against a published study, so CacaoIQ does not call them validated.
The analyses it runs
You pick a workflow by what you want to find out. The method and software behind each one are in the technical notes, and how-to articles are in the help center.
Gene discovery
The full run shown above, from raw phenotypes to candidate genes, with genomic h² from the markers.
Phenotypic modeling
Spatial or design-corrected BLUPs, repeatability or clonal H², and trait correlations. For factorial crosses, GCA and SCA with h² and d².
Trait mapping
GWAS from trait values you have already adjusted (BLUEs or BLUPs), and QTL mapping when both parents are genotyped (CacaoIQ builds the map).
Diversity panels
Population structure in a germplasm collection, with no cross or parents needed.
Your data stays on your computer.
The Matina 1-6 v1.1 gene models and their UniProt and InterPro annotations are bundled with the app.
- Systems
- Windows (64-bit); macOS 13 or later on Apple silicon
- Memory
- 16 GB recommended, 8 GB minimum
- Disk
- At least 10 GB free: the app takes about 4.5 GB, the rest is room for results.
- Graphics
- No graphics card needed
- Install
- One download with Python, R and Java bundled. No terminal.
- Network
- None needed.
Common questions
Does the language model produce the results?
No. The statistical models produce every number, and the gene summary only quotes the bundled annotations.
What happens after the 14-day evaluation?
The evaluation license expires. The Licensing page explains what comes next.
Is CacaoIQ only for cacao?
Yes. CacaoIQ is set up for cacao: its defaults, the Matina 1-6 reference genome and the gene annotations. Products for other crops are each tailored to that crop and validated against its own published studies before release.
See it run on cacao data.
Use your own trials or public cacao data.