Daniel Trejo

Computational biologist, and data scientist for fun and profit.

[Daniel Trejo @danieltba]

@anshulkundaje Care to share these?

[volcanito @tapacaminos17]

Un cuento de borges en el que cada mundial se agregan más equipos y méxico avanza a una etapa más pero nunca llega a cuartos

[Daniel Trejo @danieltba]

@anshulkundaje World model involves a latent representation; foundation models are trained to learn the conditional distribution of the data, with or without a latent representation.

[Fabian Theis @fabian_theis]

Excited to share our RegVelo paper in Cell
https://t.co/ZAnQphaXsg

We unify RNA velocity + GRNs into one model → better OOD prediction of perturbations (e.g. gene KOs), with examples incl. neural crest KO predictions 🔬

Big thanks to W Wang, Z Hu & T Sauka-Spengler 🙏

[Daniel Trejo @danieltba]

@anshulkundaje I agree, but which form of peer review can be adopted? its impossible to keep track of the ever increasing number of pre-prints, how to know which ones to focus on for properly peer-review?

[Daniel Trejo @danieltba]

@fly51fly Seems to have some relationship and shared insights with https://t.co/JR4Ti8AsA6 , both are very eye-opening.

[Daniel Trejo @danieltba]

@leonpalafox Me paso a mi, meses desperdiciados en una migracion fallida, afecto mucho mi relacion con mi entonces jefe. Desde entonces torch y solo torch. Tengo colegas que intentaron JAX y al final regresaron a torch, el api es muy inestable.

[Daniel Trejo @danieltba]

@spence_jeffrey_ @Hakha_Most @MinetoOta @TheNikhilMilind @tgjorgjieva1 @Courtsmithrun @yuvalsim @GuyS2747 @jkpritch I have problems opening the paper in biorxiv. No abstract no pdf seem to be available.

[Daniel Trejo @danieltba]

@heyokh Siempre puedes contar con que Schmidhuber va a Schmidhubear.

[Daniel Trejo @danieltba]

@anshulkundaje Wasn't that the problem with the human brain project?

[Daniel Trejo @danieltba]

@hippopedoid @KordingLab @KenjiEricLee Isn't one of the first steps in umap is building the knn graph?

[Daniel Trejo @danieltba]

@sp_monte_carlo Like some of the work we did for GWAS studies, uses ideas of caching variables and parallelism, sadly we lacked the mathematical tools to prove correctness, page 17 on https://t.co/43F3NbsPle

[Daniel Trejo @danieltba]

@PhDemetri Not for clinical prediction, but always useful to have this on the back of your head https://t.co/y0afa25Dor

[Gioele La Manno @GioeleLaManno]

Interested in navigating #metabolism in space and time? Meet uMAIA (the Unified Mass Imaging Analyzer) - a framework developed in collaboration with @Gio_Dangel0 & @OatesLab @epflSV led by @HSchede & @Leila_Alieh
#omics #lipidtime #MALDI

https://t.co/Z0mp9P7A6J
1/N

[Bastian Grossenbacher-Rieck @Pseudomanifold]

IT pro-tip: your infrastructure cannot be attacked if you brick all devices. #Crowdstrike

In all seriousness: this incident once again demonstrates the dire state of an industry dominated by snakeoil salespeople. Against the backdrop of hyped AI-based products, no one focuses on 'boring' things like reliability, stability, or—here's a novel thought—actual value.

Despite the constant failures of 'cybersecurity' products like this, everyone loves to buy them. And sooner or later, they all get bitten. The secret is that this kind of software provides some plausible deniability. If you get hacked, it's not really your fault, right? Wrong! Your infrastructure cannot be protected by a magical genie but by diligent staff.

Invest more in people, less in hype.

[Daniel Trejo @danieltba]

@heyokh @M_Sputnik Nel

[Daniel Trejo @danieltba]

@leonpalafox What is the data supporting your assertion? I would assume it is way less than 8 but i could be wrong.

[Daniel Trejo @danieltba]

@PhDemetri I also gave up on quarto css, now i just build the skeleton in quarto and compile to pptx to do the final deck in powerpoint

[Daniel Trejo @danieltba]

@solsona Mi seguro no va a pagar mas :)