Help center

Should I upload raw measurements or BLUPs?

When to upload raw measurements and when to upload BLUPs or BLUEs you already have.

Both work, and you decide per trait. With raw measurements, CacaoIQ fits the model your trial design supports, and the heritability it reports refers to the trait as measured in the field. Values you already modeled (BLUPs, BLUEs or adjusted means) are used as they are, and the results carry the cautions described below.

How to tell CacaoIQ which is which

In the run settings, Already BLUPs/BLUEs has a box for each selected trait:

  • Ticked: the values are used as they are. No BLUP model is fitted; GWAS, QTL and heritability run on your values.
  • Unticked: the trait goes through CacaoIQ’s model, even if other traits in the same file are BLUPs.

A box starts ticked when the column name contains BLUP, BLUE, adj or adjusted, lsmean or emmean, and the screen says so. This is only a suggestion: untick it if the values are raw measurements. Changing the box updates the data check in place, because negative values are normal in BLUPs and are then no longer flagged.

Three questions for already-modeled values

For each ticked trait, CacaoIQ asks:

  1. Computed with markers or a pedigree? (GEBVs or EBVs). If yes, genomic heritability is not computed, because the relationships would be counted twice. GWAS and QTL still run, with a caution that their tests are not calibrated and false positives are possible. For mapping, use BLUEs, deregressed values or raw data instead.
  2. What are these values? BLUPs, BLUEs or adjusted means, or plain means. BLUPs carry a caution that shrinkage differs between genotypes with more and fewer records.
  3. Scale: original units, log, logit or other. The scale is shown with every result.

You can leave any question at Not sure. The values are then treated as field-model BLUPs on the original scale, and the screen says so.

What changes in the results

  • Heritability: heritability from already-modeled values is not the heritability of the trait as measured in the field, because shrinkage and earlier adjustment change the variance. Values that are already means over replicates or years give an entry-mean-like h², higher than a single-tree h². The results show this caution.
  • BLUPs or BLUEs for mapping: BLUPs are for ranking trees and clones within a trial. When values go into another analysis with its own genetic model (such as GWAS across trials), or when the data are very unbalanced, BLUEs are usually the better input.

If you upload raw measurements

CacaoIQ fits a mixed model to clean the field data: the year as a fixed effect, and the tree plus any blocks, replicates and row or column positions as random effects. It picks the model family from the trait (continuous, count, presence/absence or ordinal score). Measurements recorded as whole numbers with many different values are currently treated as counts; see Known issue: whole-number traits read as counts. For continuous traits with repeated records, its phenotype tables give both the BLUP and a BLUE column with its standard error. GWAS and QTL use the BLUP.

CacaoIQ’s own BLUPs treat trees or clones as independent (no pedigree, no markers). They are clonal or total genetic values, not breeding values. To choose parents, use GCA from a factorial trial or genomic values.

Still stuck? Write to support@scilytics.ai with your system and what you see on screen. Don’t attach data files. The first reply comes within 2 business days.