@giliapatterson @petrelharp @pastramimachine and I just released our neural network that infers spatial maps of dispersal and population density using SNPs:.

RT @molecology: Out in MER!.𝑚𝑎𝑝𝑁𝑁: a SNP-based inference tool for estimating geographic maps of population density and dispersal, by @chris….

RT @min_jiseon: 1/ Individual-based simulations are a great way to model populations with less abstraction and more….

We demonstrated mapNN on North American grey wolf data from Schweizer et al. Estimated dispersal distances increased with latitude, consistent with reports that wolf range size increases with latitude. Density was too high in Eastern Canada, likely due to reduced sampling there.

Another selling point is mapNN uses SNPs, whereas MAPS (fourth column, above) requires IBD blocks so it doesn't work well with the 5,000 SNPs we gave it in this experiment.

One selling point for mapNN is it tries to distinguish dispersal and density, whereas FEEMS (third column, above) estimates a single, "combined" map.

mapNN (name of the program) trains on SLiMulations customized for the study species (geographic range, sample size, etc), and outputs pretty maps of demographic parameters. This figure shows ground truth maps in first column, and mapNN output in second column:

RT @peter_innes1: Our paper on alternative splicing evolution in Great Sand Dunes sunflower ecotypes is now published in @HeredityJournal !….

RT @Johri_Parul: Kavita Jain and I are organizing a symposium entitled "Genomic diversity in nonequilibrium populations" in the AsiaEvo Con….

"Convolvulator" is credited to Nate Pope.

The new program is available on PyPI:

We're calling the new architecture a 'pairwise convolutional network' (aka, "the convolvulator").

The new model extracts summary information from pairs of genotypes at a time. And the geographic distance between a pair is also tacked onto the extracted features. All pairs are then combined to estimate the dispersal parameter. The new model is much more accurate.

Our previous method, disperseNN, did not take advantage of available geographic information, but relied only on genetic variation for estimating dispersal.

Check out my preprint with @pastramimachine, a brief update for our deep learning model that estimates dispersal rate from SNPs:

RT @ramencult: i wrote a paper about using machine learning to identify patterns of co-dispersal in Anopheles and Plasmodium. https://t.co/….

RT @drschield: Hello #JMIH23! I am recruiting PhD students to my new lab @uvabio starting Fall '24. I won't be at JMIH to meet in person, b….

My #Evol2023 t-shirt is the smallest shirt I’ve ever tried to put on. Anyone want an XS (tag says medium)?.

RT @evolauterbur: Very pleased to release FlexSweep and some new summary stats, designed for versatility in detecting selective sweeps of d….

Beautiful location for EVO-WIBO conference: Fort Worden Historical State Park, WA