Challenges to using generative AI and machine learning by Maria Brbic:
Biology needs methods that can generalise across different tasks, different domains, different modalities, and that can discover new phenomena #BC2basel

#SIB Early Career Award very very deserved to Maria Brbic https://people.epfl.ch/maria.brbic, who is an outstanding #SingleCell #machinelearning scientist with a great commitment to equality and inclusivity in science and to #OpenSource #BC2basel #EDI

👏 #SIB PhD paper award goes to Viktor Petukhov for Cell segmentation in imaging-based spatial transcriptomics, with open code at https://github.com/ZhuangLab/MERFISH_analysis https://doi.org/10.1038/s41587-021-01044-w #BC2basel

Best part of #BC2basel: the #SIB bioinformatics awards, highlighting PhD paper, Early career, and Innovative resource https://www.sib.swiss/bioinformatics-awards #bioinformatics #ComputationalBiology

I've seen a lot about aligning spatial functional genomics in this session at #BC2basel, which makes me wonder how much they could be applied to aligning data between (close) species and not only between experiments. #EvoDevo

What are the clinically meaningful ways of looking at tumor RNA-Seq? Full thread here:
https://mstdn.science/@idavydov/111005919978862122

Poster pitch at 9:45 on Wednesday and come to the poster A6 at 14:15 (Wed) #bc2basel

Moscot + CellRank2 (https://www.biorxiv.org/content/10.1101/2023.07.19.549685v1.abstract) allow to recover brain development driver genes in mouse embryos, including regional specific drivers e.g Neurod2 #BC2basel

Marius Lange #BC2basel: increasing complexity of #SingleCell data leads to problems of temporal, spatial and spatio-temporal mapping. The solution proposed is optimal transport applications, implemented in Moscot:
https://moscot.readthedocs.io/en/latest/
https://www.biorxiv.org/content/10.1101/2023.05.11.540374v2 #SpatialTranscriptomics #scRNAseq

Excellent conclusions of Jean Fan #BC2basel on #machinelearning applied to #scRNAseq #SpatialTranscriptomics

Impressive work by Jean Fan's lab #BC2basel: aligning different spatial transcriptomics results, not only from different experiments but different protocols, e.g. MERFISH - Visium. Combining with #scRNAseq clustering and her lab's spot deconvolution allows to compare directly gene expression between these different experiments, https://www.biorxiv.org/content/10.1101/2023.04.11.534630v2.abstract #SpatialTranscriptomics

Jean Fan: "We can always machine learning to any data and it will always give us an answer" but we must be aware of the differences between data (single-cell resolution or not, full transcriptome or not…) to use the appropriate approach and obtain relevant results. #BC2basel

Jean Fan #BC2basel: Single-cell methods assume a full transcriptome access, whereas spatial transcriptomics only access a skewed subset of genes, which impacts normalisation and downstream results https://www.biorxiv.org/content/10.1101/2023.08.30.555624v1

Jean Fan https://jean.fan is the invited speaker for the #BC2basel session on Machine learning algorithms for advancing spatial biology. Very active on Youtube and X, but not here 😞

Yesterday, I listened to @ElisabethBik at #bc2basel. As always, quite revealing and eye-opening, even after having heard similar version online.

At one point, she made reference to impact. IMHO, she has had way more impact on science than most scientists.

Another detail that stuck and that impressed me: her work lead to 1000 retractions!

@ElisabethBik alerts about AI generated fake images. I just generated "mastodon post about keynote talk by Elizabeth Bik" 😅 #BC2basel

Of 782 papers reported to journals, no action was taken for 65% after 5 years! Only 7% retracted. Thus @ElisabethBik started posting publicly issues on @PubPeer: >7k papers reported to PubPeer, 1000 retractions (exactly as of today!). You can see problems with papers you're reading with a PubPeer plugin https://pubpeer.com/static/extensions #BC2basel #PubPeer

@ElisabethBik feels that some cases should be "5 min retractions" and not need a lengthy investigation, e.g. with multiple duplications with obvious manipulation #BC2basel

@ElisabethBik makes the audience guess image duplications, and I can now confirm that I'm really bad at this… #BC2basel

I like how @ElisabethBik shows examples of misconduct just with their DOI, not to point fingers at any specific authors (nor depending on DOIs, specific journals). #BC2basel

@ElisabethBik: "Science:
• builds upon science
• is about finding the truth
• is built on trust
• is not immune to fraud"
#BC2basel