With less than a 96well plate worth of assayed variants, our semi-supervised eUniRep method constructs accurate virtual fitness landscapes and screens ten million protein sequences with in silico directed evolution
. (2/7)pic.twitter.com/PkQ5NwuzGK
U tweetove putem weba ili aplikacija drugih proizvođača možete dodati podatke o lokaciji, kao što su grad ili točna lokacija. Povijest lokacija tweetova uvijek možete izbrisati. Saznajte više
With less than a 96well plate worth of assayed variants, our semi-supervised eUniRep method constructs accurate virtual fitness landscapes and screens ten million protein sequences with in silico directed evolution
. (2/7)pic.twitter.com/PkQ5NwuzGK
With just hours of GPU computing and 24 assayed mutants of wild GFP
as training data, we generate much brighter diverse fluorescent proteins, some practically as bright as sfGFP - a fruit of a multi-year effort in high-throughput protein engineering. (3/7)pic.twitter.com/zEKzH2qxUH
eUniRep also engineers enzymes and leverages epistasis! Despite training on just single mutants of beta lactamase, we generate many >WT epistatically non-trivial variants. (4/7)pic.twitter.com/WLPB2mHGTk
Even without any experimental data, eUniRep encodes protein function from raw amino acid sequences alone as the representation’s first principal component. This is likely why eUniRep generalizes so well - maybe even less training data is needed! (5/7)pic.twitter.com/SdGTfvjHHa
Low-N protein engineering with eUniRep+in silico evolution enables engineers to screen 10s of millions of variants even if assaying the target function is very resource intensive, as is often the case in drug development, agriculture, and industrial production. (6/7)
Stay tuned for code and more! (7/7) #proteinengineering #MachineLearning #UniRep #eUniRep
At first I thought this was about making proteins with less nitrogen
I don’t know how I missed that before
Are you familiar with ULMFiT? Your self-supervised pre-training and fine-tuning steps follow an identical protocol ULMFiT: https://arxiv.org/pdf/1801.06146.pdf … ULMFiT applied to Protein Sequences: https://www.biorxiv.org/content/10.1101/704874v1.full.pdf …
Thanks! Our evotuning generally follows our previous work https://bit.ly/2RtuJHK , different from ULMFIT e.g. we use fixed representation with linear top model for supervised tasks.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.