iiinova — where your digital self keeps evolving

Alvaro Sanchez @asanchezlab.bsky.social · 30/08/2026

e.g. this is the global epistasis data for a dataset measuring microbiome effect on several life-history traits of Drosophila, by alison gould et al. Frankly I have no idea where I'd start if I wanted to create a CRM for this. Maybe Alison, Will et al have figured it out by now but it's not obvious!

000

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

The most surprising consequence is that presence/absence remained highly predictive of community function even after 10days of population dynamics. Communities moved far from their initial abundances, yet knowing which species were present at the start predicts community biomass at equilibrium!

111

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

This leads to a final hypothesis: if the dominant biomass mode is set by species yield, it should remain stable as species abundances change. We tested this over 10 days of serial passaging. Despite large shifts in abundance, the dominant mode barely moved and stayed aligned with yield.

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

This is of course the classic "non-linear transformation of a latent, additive variable" that is the hallmark of global epistasis in genetics, only now demonstrated for ecological communities, and calculated explicitly from independently measurable parameters.

130

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

Importantly, this collapse is not just descriptive. Once the leading mode (u_1) is known, together with its eigenvalue and the additive effects, we can actually calculate the one-dimensional curve that organizes the landscape. With two modes, the same idea gives the full two-dimensional surface.

200

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

Consumer-Resource models gave us a clear prediction: the dominant collective mode that makes community-function landscapes approximately 1-dimensional should be approx. the same as a vector of species traits, formed by the metabolic efficiencies, or yields, of each species.

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

This is interesting but provides little ecological insight. What do these collective modes actually represent biologically? For biomass, we turned to consumer-resource models, where species compete for shared resources and their underlying traits are explicitly defined.

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

This simple organization has a major consequence: species’ effects on community function do not vary independently across backgrounds. Instead, they rise and fall together. As a result, the full community-function landscape becomes a function of just one or two ecological collective modes.

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

To test this prediction, we built complete community-function landscapes containing every combination of five bacterial species, measuring both biomass and pyoverdine production. We then repeated the entire experiment across 31 different resource environments.

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

The answer is remarkably simple: if a species follows a good FEE, its pairwise interactions scale with its additive effect, its three-way interactions with the corresponding pairwise ones, and so on across all orders.

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

A clue comes from the previous observation of Functional Effect Equations (FEEs): the effect of adding a species often depends linearly on the function of the community it joins. But that effect itself emerges from interactions at all orders. What does a strong FEE tell us about those interactions?

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

New preprint from the lab: The Latent Simplicity of Microbial Ecological Interactions www.biorxiv.org/content/10.6... We're excited about this one. We find that high-order microbial interactions often obey simple linear laws making microbial communities far more predictable than one might expect

37032

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

The models gave us a clear prediction: the dominant collective mode that makes community-function landscapes approximately 1-dimensional should be approx. the same as a vector of species traits, formed by the metabolic efficiencies, or yields, of each species.

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

We then asked what this implies for the underlying interactions. The answer is remarkably simple: if a species follows a good FEE, its pairwise interactions scale with its additive effect, its three-way interactions with the corresponding pairwise ones, and so on across all orders.

100

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

110

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

4277

Alvaro Sanchez @asanchezlab.bsky.social · 21/08/2026

New preprint from the lab: The Latent Simplicity of Microbial Ecological Interactions www.biorxiv.org/content/10.6... We're excited about this one. We find that high-order microbial interactions often obey simple linear laws, making microbial communities far more predictable than one might expect.

12710

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

Our results suggest that designing better growth media (for biotech, bioprocessing, synthetic biology) doesn't require exhaustively screening every mixture. A relatively small, well-chosen set of experiments can teach a model the whole landscape.

110

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

This has practical consequences: simple regression models (just additive + pairwise terms) trained on a sparse training set of environments predict growth in brand-new, untested resource mixtures with R² > 0.9. This occurred systematically and for all 7 strains tested.

110

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

We checked directly: across all 255 environments, additive effects + pairwise interactions explain >90% of the variance in growth for every single strain. High-order interactions exist, but they're a rounding error.

110

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

Interestingly, the slope of that regression is explained almost entirely by pairwise interactions between the added resource and the others already present. No need to invoke complex higher-order interactions. This suggests HOIs amongst resources should contribute little to bacterial growth.

110

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

A fair fraction of the fitness effect of adding a resource is explained by a simple linear regression on the fitness of the background mixture we add it to, i.e. that resource's "Fitness Effect Equation". This is similar to how global epistasis explains part of the fitness effect of a new mutation.

120

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

We asked whether these sign changes were idiosyncratic, caused by specific nutrient intxns, or whether they followed a global pattern analogous to global epistasis in genetics. Indeed, for most resources, the sign reversal effect was partly explained by the fitness on the background resource mixture

110

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

Even resources that support growth when they were alone could hurt growth when added to certain mixtures. For instance, acetate alone helps P. atacamensis grow. But adding acetate reduced fitness in 32/128 cases. Context is everything.

130

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

First, we found that even though every C source was present at a low concentration and, *on average*, growth increased as we added additional nutrients, almost every resource could have detrimental effects on growth in some mixtures, depending on which other resources were in present in the mix.

120

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

We used a full factorial design with 8 carbon sources (sugars, sugar alcohols, organic acids), all 255 combinations, 7 strains (5 Pseudomonas + Salmonella + Serratia), under carbon-limiting conditions. This allows us to directly measure interactions at every order, not just pairs.

110

Alvaro Sanchez @asanchezlab.bsky.social · 01/07/2026

Optimizing culture media is a major goal in microbiology & biotech. A challenge is that nutrients don't necessarily just add up their effects on growth: they can potentially interact. With as few as 20 resources (a very conservative number) there are >1M possible interactions.

110

Alvaro Sanchez @asanchezlab.bsky.social · 09/03/2026

Note that these two factors are purely statistical and do not speak of the underlying biology. As the number of species increases, the relative contribution of higher-order interactions to the total variance will eventually grow, simply because there are many more of them.

120

Alvaro Sanchez @asanchezlab.bsky.social · 09/03/2026

Part of this pattern is structural. Two factors shape the variance spectrum: how many interactions exist at each order, & a geometric factor that suppresses higher orders as the number of measurements needed to estimate them grows. Together they naturally bias landscapes toward low-order structure.

120

Alvaro Sanchez @asanchezlab.bsky.social · 09/03/2026

To analyze the global structure of the landscape, we decomposed the total functional variance by interaction order. Consistent with previous work, first- and second-order interactions explain most of the variance, while higher orders contribute very little.

110

Alvaro Sanchez @asanchezlab.bsky.social · 09/03/2026

In this framework, interactions appear as epistasis. If the effect of adding one species depends on whether another species is present, we have an interaction. Higher-order interactions involve three or more species simultaneously.

120

Alvaro Sanchez @asanchezlab.bsky.social · 09/03/2026

We study community-function landscapes, where each species may be either present or absent, and each community composition maps to a measurable function (e.g. total biomass). This defines a high-dimensional map from composition → ecosystem function, inspired by fitness landscapes in genetics.

140

Alvaro Sanchez @asanchezlab.bsky.social · 16/12/2023

New call for postdoc fellowships from the 🇪🇸 govt. My brilliant colleague Cristina Vieitez and I are happy to jointly sponsor candidates who'd like to join us in beautiful Salamanca & develop methods for characterizing and optimizing microbial consortia. More info ⬇️. www.aei.gob.es/convocatoria....

0810