Kumar Mainali

13 secret websites powered by AI to finish hours of work in seconds (Don't lose them):

Exciting news! Our AI system for predicting wetlands with very high accuracy has been published in Science of The Total Environment and featured in 27+ media outlets. #AI #wetlands #prediction phys.org/news/2023-01-a…

CooccurrenceAffinity: An R package for computing a novel metric of affinity in co-occurrence data that corrects for pervasive errors in traditional indices biorxiv.org/cgi/content/sh… #bioRxiv

10 academic writing skills I wish I had learned as a PhD student (with updated links) A thread ⬇️ 🧵 for PhD students and junior scholars:

We also highlight work by @kumar_mainali and colleagues on a new metric of co-occurrence and similarity that addresses existing sensitivity issues present at more popular similarity indices (rdcu.be/cHl25). nature.com/articles/s4358…

This is really important stuff to consider if you use dissimilarity indices ever (think ordination, genetic analyses, community analyses, beta diversity, etc). Curious to see whether this will catch on! science.org/doi/full/10.11…

@vjjan91 @bio_diverse @mizoraman @PriyankaHariH @vjjan91 the paper is under open access. Please check again.

Have you ever calculated Jaccard's index? I have. Stop, and read this paper before you do so again. It proposes a new measure of co-occurrence that changes the interpretation of patterns in your data #biodiversity science.org/doi/10.1126/sc…

@Jon_Chase03 I need to sit down, get pencil and paper, and do math before I can start singing: "And now, the end is near And so I face the final curtain My friend, I'll make it clear I'll state my case, of which I am certain" Looking forward to spending time on this one!

@Jon_Chase03 @DanMcGlinn sorry about my multi-part response... I am not used to the brevity of Twitter.

Amidst a daunting array of "better" measures of co-occurrence and community similarity (e.g., beta-diversity), this one seems rather important: science.org/doi/full/10.11…

@Jon_Chase03 #7) @DanMcGlinn Articles on co-occurrence indices seem to recognize that dramatically non-null association between species pairs is actually rather common, & the point and confidence-interval estimation of the degree of non-nullity ought therefore to be of scientific importance.

@Jon_Chase03 #6) @DanMcGlinn In contrast, the log odds ratio is a parameter that is meaningful to interpret in the way the two species interact (ie they have greater probability of choosing the same sites because they are looking for some of the same things).

@Jon_Chase03 #5) @DanMcGlinn Centering and scaling, ie standardizing, under null hypothesis is "roughly" comparable to doing p-values (also a null-hypothesis calculation), and that seems to be where Carmona and Pärtel (2020) stop.

@Jon_Chase03 #4) @DanMcGlinn There is a considerable difference betn analysis done under a legitimate interpretable alternative hypothesis and that done under null hypothesis (hypergeometric distribution of co-occurrences applicable when site occupancy is truly independent across species).

@DanMcGlinn @Jon_Chase03 @R_you_cereal @C_PCarmona We believe that this won't make much difference in larger-sample 2x2 tables, but definitely could in smaller tables. It's an issue of sensitivity analysis, which could be addressed in Bayesian fashion by putting a prior on the row-column totals where constraints should be soft.

@kumar_mainali @Jon_Chase03 @R_you_cereal @C_PCarmona I'm also curious about relaxing the marginal constrains. It is interesting to me that subtle and somewhat arbitrary decisions (hard vs soft constraints) about holding richness or abundance fixed can strongly change the null model and thus these kinds of metrics.

@Alex_Smith_Ants Totally agree on the confusion our choice of name has caused. We should do a better job in naming a metric.

@LeffUfrgs The first two pages summarize the challenges with the prior methods. Fig 1 and 2 show how various indices and alpha map to CDF.

@Jon_Chase03 #2) @DanMcGlinn First, for a given scenario of prevalence, the distribution is not symmetric. This means that values equidistant from the center of the null in opposite directions (e.g., 2 versus −2) indicate different strengths of positive and negative association.

@Jon_Chase03 #1) @DanMcGlinn Compared to raw Jaccard, its standardization (Keil 2019, Ecosphere) is much better. We show the standardized Jaccard index correctly centers the value of zero at the center of null (section S2 and fig. S1, first column).... However, it still presents two problems.