November 07, 2018 at 08:13PM

November 07, 2018 at 08:17PM

Gamze Gursoy (from @MarkGerstein lab) describes how to strip out info from RNAseq BAM files (=> pBAM) to still allow expression analysis but suppress "data leakage" and its ability to be used to deidentify subjects. #biodata18

November 08, 2018 at 09:03AM

@mike_schatz

November 08, 2018 at 09:05AM

November 08, 2018 at 09:13AM

November 08, 2018 at 09:19AM

Anna Goldenberg from University of Toronto: no way to necessarily tell which model offers best explanation if predictive capability is similar - can consider generative models instead to help with that. #biodata18

November 08, 2018 at 09:24AM

November 08, 2018 at 09:36AM

@MarkGerstein

.@MPloenzke shows nice movies of NN training to find seq. motifs (https://t.co/rnenBuwvg0). After some interpretation, the 1st-layer convolutional filter in the network can learn motifs. He shows how one can simplify the process & use NNs to find motifs directly. #BioData18

November 08, 2018 at 09:39AM

@MPloenzke MH's slides are at https://t.co/rVGdA5FESm, code at https://t.co/LJeAwbMSuU, paper at https://t.co/ZGRyDF6HxK #biodata18

November 08, 2018 at 09:52AM

#biodata18 @jmschreiber91's pre-print for avocado. its a tool to accurately predict those epigenetic peaks in cell differentiation :
https://t.co/fL8GTLWIqa

November 08, 2018 at 09:54AM

@MarkGerstein

.@jmschreiber91 describes Avocado, a deep neural network tensor factorization method. It learns a latent, low-dimensional representation of the human epigenome, which can be used for accurate chromatin imputation & predicting gene-expression & FIREs #BioData18

November 08, 2018 at 09:57AM

@seandavis12

November 08, 2018 at 10:03AM

November 08, 2018 at 10:08AM

#biodata18 @avshrikumar https://t.co/OsRnTqHJ2d 's secret sauce - integrated gradients are used to get around saturation for deep learning in genomics. also, check this out:

https://t.co/lEJ2KdsQNQ

November 08, 2018 at 10:41AM

@MVickySchneider

November 08, 2018 at 10:43AM

November 08, 2018 at 11:06AM

Enjoying twitter following dual conferences #sacb2018 #BioData18 — I feel like this calls for a comp bio nerd fight for the best meeting

November 08, 2018 at 11:11AM

@michaelhoffman

November 08, 2018 at 11:12AM

compartmap @Bioconductor #rstats 📦 tool by @biobenkj @timtriche extends group level compartment by @JayPykw @KasperDHansen to single cells/samples

#biodata18

November 08, 2018 at 11:16AM

#biodata18 https://t.co/la555HGt56 excels in getting the distances between the clusters. it supports a wide variety of distance functions, including non-metric distance functions such as cosine distance and correlation distance @biobenkj

November 08, 2018 at 11:16AM

@michaelhoffman

BJ: We took same dataset and used semi-supervised UMAP and compartmap. Sorted host and leukemic fraction cluster together [wut] #biodata18

November 08, 2018 at 11:17AM

@MarkGerstein

.@biobenkj refs to UMAP (Uniform Manifold Approximation & Projection for Dimension Reduction) https://t.co/NTXl4nbdOt QT:"UMAP algorithm is competitive with t-SNE for visualization...arguably preserves more of the global structure with superior run time performance." #BioData18

November 08, 2018 at 11:22AM

@michaelhoffman

And now for something completely different. Now: Alex White: Modeling and visualizing ecological structure across the tropical-temperate divide. "We're not going to talk about humans. Or even molecular data." #biodata18

November 08, 2018 at 11:24AM

@SIDataScience

Postdoc Alex White is talking about his dissertation research modeling and visualizing ecological structure. R package available here: https://t.co/026c3qjvcD #BioData18

November 08, 2018 at 11:26AM

@michaelhoffman

November 08, 2018 at 11:29AM

@michaelhoffman

AW: Want to use a probabilistic model to infer biotas. Don't want to force discrete biotas. Use a mixed membership model. 9518 global bird range distributions into biotas. cc @ProBirdRights #biodata18

November 08, 2018 at 11:34AM

@michaelhoffman

AW: Local example in the Himalayas. Collected from 38 local forest (5 ha) surveys of breeding birds. 620 breeding species. [can any ecology-adjacent folks explain the significance of *breeding* here? What's a non-breeding bird species? cc @stepheniwright @jrossibarra] #biodata18

November 08, 2018 at 11:34AM

@MarkGerstein

AW at #BioData18 describes a simple mixed membership model (determined from matrix factorization) to describe observed species locations in terms of inferred biotas. Model recapitulates Wallace's partitions.

November 08, 2018 at 11:37AM

@michaelhoffman

AW: Regional biomes suggested by Wallace have strong support. Species and lineages do seem to be constrained by freezing. Temperate and tropical regions can have decoupled histories. R package available. #biodata18 https://t.co/6UC0Zgtmo5

November 08, 2018 at 11:40AM

@michaelhoffman

Q: How can you share data but prevent animal poachers from using it? AW: Himalaya data is available in our R package. Lots of other data out there already including from IUCN. #biodata18

November 08, 2018 at 11:41AM

#biodata18 @pbelardo freezing lines are correlating to location of species and region biomes. also, the himalayan have ~620 bird species https://t.co/NEL3IOvsGL

November 08, 2018 at 11:45AM

@michaelhoffman

Now: Genevieve Stein-O'Brien (@GenesOfEve): Decomposing cell identity for transfer learning across platforms, tissues and species. With @FertigLab @loyalgoff #biodata18

November 08, 2018 at 11:54AM

@michaelhoffman

GS preprint describing transfer learning via projectR #biodata18 https://t.co/BPwqq62iWk https://t.co/jMs6lbJ5qD https://t.co/MBxX4pUCx9

November 08, 2018 at 11:56AM

@michaelhoffman

GS: ProjectR quantifies pattern use in different cell types in independent data. Application to Tabula Muris data #biodata18 https://t.co/11cmYhbNUT

November 08, 2018 at 11:56AM

@MVickySchneider

ProjectR: transfer of discrete type cell signatures from CoGAPS (a Bayesian MCMC matrix factorization algorithm, that links to gene set statistic methods to infer biological process activity). #biodata18 https://t.co/OKiOB84hWx https://t.co/l4uYgnOF7F https://t.co/ECFfR8V5T8

November 08, 2018 at 11:59AM

@MarkGerstein

.@GenesOfEve (from @FertigLab) at #BioData18 describes scCoGAPS (https://t.co/ctoUy1rGbY) Program adapts a sparsity constraint to better use NMF for single-cell data. Cell-type signatures from this program go into ProjectR (https://t.co/vbx4BnTHya) for further investigation

November 08, 2018 at 01:46PM

@nothingclever

In the “future of biological data science” panel, @mike_schatz points out the importance of incorporating new technologies (e.g. long reads) to get a more accurate genotype for genotype/phenotype data. #biodata18

November 08, 2018 at 01:47PM

From the audience: I'm starting to look into wearables because lots of people have them already. Problem is, most of this data is stored in companies, not dbGaP. How can we get access to this type of data?

@aphillippy: how can we get data from corporate entities?

#biodata18

November 08, 2018 at 01:49PM

Answer by @DrAnneCarpenter: can't comment on buying the data from a company, but remember that a user also owns it and could voluntarily submit it

Hm... I think that @fitbit doesn't let you download your data. But I haven't checked in years so I don't know.

#biodata18

November 08, 2018 at 01:51PM

@DrMeghanFerrall

November 08, 2018 at 02:06PM

@MarkGerstein

At the #BioData18 panel, @MSchatz discusses a data science pyramid (from his previous article on biological data sciences in Genome Research https://t.co/PJ8hWOegvM)

November 08, 2018 at 02:18PM

@JasonWilliamsNY

November 08, 2018 at 02:26PM

@cshperspectives

Question about genetic data privacy at #biodata18. Is it impossible to secure? @aphillippy points out all you need is a hair [insert Gattaca ref]. Solution is surely better protection legislation (extend GINA, etc) rather than privacy tech that can be circumvented.

November 08, 2018 at 02:38PM

@DrAnneCarpenter introducing the inverse of Moore's law, Eroom's law - which describes the skyrocketing costs of developing a new drug (also on a log scale) #BIODATA18

November 08, 2018 at 02:39PM

@michaelhoffman

AC: Eroom's law: Moore's law backwards. Skyrocketing cost to develop a new drug. In 1950s, $1B = 30 new drugs. Today $1B gives you 1/3 of a #biodata18

November 08, 2018 at 02:41PM

@micknudsen

November 08, 2018 at 02:43PM

@MirianTsuchiya

November 08, 2018 at 02:47PM

#biodata18 @DrAnneCarpenter yay humanity - over the course of 10 years we've gone from knowing almost nothing about disease to successfully completing tons of gwas studies

ftp://ftp.ebi.ac.uk/pub/databases/gwas/timeseries/current/GWAS_Catalog_diagram.svg

November 08, 2018 at 02:49PM

November 08, 2018 at 02:56PM

Go to https://t.co/APKzkfiUWc to see more examples of how the software has been used including resulting clinical trials #BIODATA18

November 08, 2018 at 03:03PM

November 08, 2018 at 03:05PM

@michaelhoffman

November 08, 2018 at 03:20PM

@MarkGerstein

At #BioData18, @DrAnneCarpenter cites "Systematic morphological profiling of human gene & allele function via Cell Painting," https://t.co/37hN26VYGn Over-expressing genes, doing cell profiling, & then finding genes sensibly cluster based on cellular morphological similarity

November 08, 2018 at 07:34PM

@mike_schatz

November 08, 2018 at 07:39PM

November 08, 2018 at 07:50PM

@michaelhoffman

Now: Ziga Avsec (@avsecz): @KipoiZoo: Accelerating the community exchange and reuse of predictive models for regulatory genomics. #biodata18

November 08, 2018 at 07:50PM

November 08, 2018 at 07:51PM

@michaelhoffman

November 08, 2018 at 08:03PM

Work done by ⁦@msimbirs⁩ @DNAnexus: Kipoi variant scoring as a DNAnexus applet, so we can apply Kipoi's deep learning regulatory models with data in cloud. #BIODATA18 https://t.co/86J8ITOAzr

November 08, 2018 at 08:09PM

@MarkGerstein

At #biodata18, @avsecz cites "Kipoi: accelerating the community exchange and reuse of predictive models for genomics," https://t.co/4o0Gdx0jB3 Points out that FAIR applies to predictive models as well as to data

November 08, 2018 at 08:16PM

@RobAboukhalil on Web Assembly. (https://t.co/GbEWHJVMm5) Not to be confused with genome assembly -- this is the CS secret sauce that allows tools like https://t.co/KhSbLi3D8f run extra fast in browser #biodata18

November 08, 2018 at 08:29PM

@michaelhoffman

Now: Anton Nekrutenko (@nekrut): Can UI handle 10ˣ datasets where x > 3? Yes, it can! [first title that requires use of notation] #biodata18

November 08, 2018 at 08:34PM

@michaelhoffman

November 08, 2018 at 08:44PM

@michaelhoffman

November 08, 2018 at 09:06PM

@michaelhoffman

November 08, 2018 at 09:07PM

@MarkGerstein

November 08, 2018 at 09:15PM

@SherlockpHolmes SH: There is a parallel between topic modeling in NLP and microbial species in communities using Latent Dirichlet Allocation #biodata18

November 08, 2018 at 09:18PM

@michaelhoffman

Now: Fabio Navarro: Developing a comprehensive resource of the human body epigenome with a new integrated annotation of tissue-specific regulatory elements and networks. With @MarkGerstein #biodata18

November 08, 2018 at 09:22PM

@michaelhoffman

SH: Unique shape associated with histone signals flanking active enhancers identified through STARR-seq #biodata18 https://t.co/DhPgMUZ8PI

November 08, 2018 at 09:22PM

@michaelhoffman

November 08, 2018 at 09:27PM

@michaelhoffman

November 08, 2018 at 09:30PM

@michaelhoffman

@MarkGerstein @FNavarroBioInfo @Mengting_Gu @bornalibran No big deal, these errors just have a way of... propagating. Thanks!

November 08, 2018 at 09:32PM

November 08, 2018 at 10:29PM

@Mengting_Gu

@michaelhoffman @MarkGerstein @FNavarroBioInfo @bornalibran Sorry about this... I must have picked it from somewhere without carefully checking it. Thanks for pointing it out! We will make sure to fix it in the final version.