Evolving notes, images and sounds by Luis Apiolaza

Category: breeding (Page 1 of 3)

Do you remember your first time?

You were nervous. Would they like it as much as you did? Would you make the cut? Your first manuscript as a senior author tends to be a memorable experience. On one side, you have been working a long time, coming to terms with the problem, learning, building models, polishing the words [insert a few iterations here] until you submit the manuscript. It is a hopeful act.

Do you remember the feeling of the first acceptance? Your work was judged good enough to be published in that special journal, the one you like. The one were so and so, the authors you admire, published their work. Later you’ll understand that there are diminishing returns, so your tenth article will not provoke the same reaction, and your fiftieth article… you get the idea.

Do you remember your first rejection? Was it just a “desk-rejection”, wrong journal, no big deal? Or was it a “we hate the manuscript, what a turd”? This one can hurt, but there are diminishing returns too: your tenth rejection is more like “meh, what do they know?”.

Both the acceptances and rejections are of that particular piece of work. They are not about you, although some referees (typically referee No2) sometimes manage to make it feel personal. You are not a better or worse person because of the comments of a random set of referees. It is good to remember that a different sample of referees could have told you something very different about the manuscript.

I do remember the first acceptance; I barely remember the first rejection. I do look at those experiences with older eyes, thinking that in both cases I would write the manuscript very differently today.

Sometimes we want more, sometimes we want less

I am running analyses for a new article with my colleague Clemens Altaner (a smart cookie), reprocessing old samples to get resin data. This got me thinking on types of traits, as in there are “we always want more” traits, like stem volume in trees, or yield per ha for many agricultural crops. Assuming there were no trade-offs (like reducing quality) we always want more stem volume.

There are also traits with technical thresholds, like wood stiffness grades, or fruit quality grades, where there is a stepwise value function. There is an extra payoff when reaching a new grade, followed by a plateau, when extra expression of the trait is worth nothing until… we get to the next step/grade.

And there are traits which are highly dependent on the end-product, like resin content or heartwood content. If you want to grow a crop for resin production (like some pines in China) you want as much resin as possible. However, if you are interested in solid wood, resin and resin canals are an annoyance, as they reduce the wood grade. A similar situation occurs with heartwood. It’s great to have heartwood if I want more durability, but it may affect processing (including preservative injection) if we want the wood for other uses.

For the current analysis and end-use we want lower values of the traits, but things can change if we move countries or industries.

I would like to know if you can give me examples of this three types of traits (or if there are more types) for your species/end-product. It’s always handy to have non-forestry examples for class.

All that glitters (still) is not GxE interaction

Five months ago I was saying all that glitters is not GxE interaction, putting forward the idea that, at least in forestry, some of the reported GxE interaction was not interaction at all but poor connectedness.

Today Forest Ecology and Management published the work by Duncan McLean (our PhD student, congratulations!), Jaroslav Klápště, Mark Paget and myself (Open Access article). We have eight well-replicated and connected trials, SNP-based pedigree, covering a range of environmental conditions with little signs of interaction.

For wood basic density, a typically highly heritable & low interacting trait, out of the 28 cross-site genetic correlations (Type B in forestry parlance) the lowest value was 0.86. For stem diameter at breast height, a typically low heritability & high interacting trait, 20/28 cross-site correlations were greater or equal to 0.7, often used to define interacting sites. The lowest correlations came from a single South Island site (lowest value was 0.59).

Trees live for a long time and tree breeding programmes are long-lived endeavours. Our genetic evaluations often include trees from many selection series, with many trials connected by tenuous relationships. There are publications that show we only need small connections to compare genotypes across trials: the predicted error variance of the comparison is OK. We need to remember though, that those comparisons are assuming we know the true genetic correlations across sites (as in animal breeding); instead, we are estimating them from data. Data coming from poorly connected trials, subject to biases.

Now imagine we start explaining the estimated GxE interaction, Are we explaining real interaction or just estimation noise? We still have to continue revisiting this topic, which I hope it leads to updating the genetic evaluation system.

Genetic correlations for DBH and wood density under three genetic models.

Not a miracle worker

—The genetic correlations are very high. Can you check them?

The additive genetic variance for one of the traits was essentially zero, which pushed the correlation to get stuck at 0.99. Looking with more attention at the results of the univariate analyses, showed that the experimental design features (replicates, incomplete blocks) were also close to zero.

A cross tabulation showed that some of the replicates had only a single observation (and many of the incomplete blocks had none), although the families had five observations each.

More importantly, and easier, checking the size of the dataset showed 150 observations. One hundred and fifty observations to estimate the degree of genetic control and associations. Way too small!

The small sample size was not done on purpose; it was just left overs from another project that we are sending for publication soon. Small leftovers, small sample size; enough for a pilot study at the phenotypic level, but not for genetic parameters. There are no miracles in data analysis.

A few things that I expect to find in a well-functioning breeding programme

A good understanding of the biological traits that have an effect on profit. Can we identify new, more efficient selection criteria?

A clear awareness of trade-offs between: traits, genetic evaluation options, deployment systems, etc. Trade-offs are everywhere in breeding programmes.

A database system that contains all the data.

A well-documented genetic evaluation system: we can rerun the evaluation and get exactly the same results. The system can be developed in-house or can use commercial software (like asreml, SAS, Bolt, etc) but the code must be available.

Reproductive biology: essential to sort out the best deployment.

Figured/figuring out what are the main environmental drivers affecting performance. It is easy for some species, very difficult for others.

Personnel continuity BUT the programme can survive the proverbial bus running over the breeder.

« Older posts

© 2024 Palimpsest

Theme by Anders NorenUp ↑