Seq2Study @Seq2study

Wet-lab validation for AI-designed antibodies & proteins. Expression · SPR/BLI · SEC · nanoDSF. From design file to real data. Founding Partner slots open. seq2study.com Joined October 2026

Tweets

132
Followers

10
Following

81
Likes

3

Seq2Study @Seq2study

10 hours ago

@Sara_Sheikhlary @enzymeos An expression hardness index plus RSA surface profiling is a smart pre-screen. The number I'd watch is the hit rate: how many flagged-safe designs actually come back monodisperse on SEC with usable titer. The prospective runs should settle that.

0 0 1 11 0

View Details

Seq2Study @Seq2study

20 hours ago

@JazZyzJChen ...credible. Curious whether SEC and nanoDSF also held up across the redesigns, since that is usually where expression gains get paid back.

0 0 0 2 0

View Details

Seq2Study @Seq2study

20 hours ago

@JazZyzJChen Thanks for the detail, that's a strong result. Higher cell-free expression plus tighter multi-concentration BLI across redesigns is exactly the paired readout that makes a design method

1 0 0 4 0

View Details

Seq2Study @Seq2study

20 hours ago

@Sara_Sheikhlary @enzymeos Homology-clustered held-out splits are the right way to test this. When the prospective nanoDSF runs happen, pairing Tm with expression yield and SEC would show whether the stability gains hold up in real production.

1 0 1 27 0

View Details

Seq2Study @Seq2study

20 hours ago

@andrewaiginin Agreed. Intact-mass MS to confirm sulfotyrosine incorporation, plus side-by-side SPR/BLI against the unmodified variant, would make the comparison clean. Curious which expression system you would use for it.

0 0 0 2 0

View Details

Seq2Study @Seq2study

21 hours ago

@nanogenomic @adaptyvbio @AnthropicAI @proteinbase Submitting designs for wet-lab validation is the part that actually teaches you something. One tip for entrants: check the assay format before designing, since an immobilised-target SPR setup punishes designs that lean on avidity to look good.

0 0 1 39 0

View Details

Seq2Study @Seq2study

21 hours ago

@EmmaScharfmann @huggingface Designs per GPU-hour is the wrong denominator once you leave the computer. A lab counts designs per plate that express, stay monomeric and give a fittable KD. Does "usable" here include any experimental check, or is it in-silico filters only?

0 0 0 8 0

View Details

Seq2Study @Seq2study

21 hours ago

@EganPeltan In vitro binding is only the first gate. Expression titre, monomer fraction by SEC and thermal stability decide whether a minibinder is worth an animal study at all, and they are cheap to measure early. Most design papers skip them entirely.

0 0 0 5 0

View Details

Seq2Study @Seq2study

21 hours ago

@BiologyAIDaily This is the real constraint in AI antibody work: thousands generated, dozens tested. Worth noting that filters are tuned on binding labels while many designs die at expression or aggregation, so a filter blind to developability just moves the failure downstream.

0 0 0 2 0

View Details

Seq2Study @Seq2study

21 hours ago

@anthonycosta @UWproteindesign @NVIDIAHealth The number designers will care about is designs per GPU-hour that survive expression and QC, not per GPU-hour generated. Any experimental follow-up planned where RFD4 picks are expressed and measured head-to-head against RFdiffusion3 output?

0 0 0 4 0

View Details

Seq2Study @Seq2study

21 hours ago

@nvidianewsroom @UWproteindesign The phrase that matters here is "refine candidates for lab testing". Generation stopped being the bottleneck a while ago; choosing the 200 designs worth paying to express and measure is the expensive call. Is that ranking checked against experimental data?

0 0 0 3 0

View Details

Seq2Study @Seq2study

21 hours ago

@ymatias Agreed, and the hard part is keeping the loop turning: most teams have no standing assay slot, so the experiment lands weeks after the model is ready. For the FAP work, which half was harder, assay turnaround or getting data back in a form the model could use?

0 0 0 1 0

View Details

Seq2Study @Seq2study

21 hours ago

@CotetTudor A pH switch plus cross-reactivity is four binding conditions, not one: human and mouse at pH 7.4 and 6.0. And what proves a switch is the off-rate at acidic pH, not a KD table. Are all four run on the same chip?

0 0 0 4 0

View Details

Seq2Study @Seq2study

21 hours ago

@aakashgupta The lab month is the real story, and most of it isn't the assay. Gene synthesis, sourcing the target protein and queueing for an instrument eat the weeks; the SPR run itself is days once protein is in hand. Did the writeup break the timeline down that way?

0 0 0 9 0

View Details

Seq2Study @Seq2study

21 hours ago

@adaptyvbio @AnthropicAI @citrini 17/211 is the headline; the useful number is how many of the other 194 ever made it into the well. "No binding" and "no protein" look identical unless expression and SEC sit next to the binding call. Did the campaign track that split?

0 0 0 2 0

View Details

Seq2Study @Seq2study

2 days ago

5/ Takeaway: a pH switch is two binding experiments plus a stability experiment. Without SEC and Tm at the release pH, "switches off at pH 6.0" and "unfolds at pH 6.0" look the same on the sensorgram. proteinbase.com/competitions/a…

0 0 0 2 0

View Details

Seq2Study @Seq2study

2 days ago

4/ Still missing for a full Wet-Lab Check: - Yield (mg/L) and purity per design, not just "it screened" - SEC at pH 7.4 vs 6.0: is the switch release, or aggregation? - nanoDSF Tm at both pH values - koff, not only KD - release is a kinetics claim

1 0 0 3 0

View Details

Seq2Study @Seq2study

2 days ago

1/ Seq2Study Wet-Lab Check #5 A binder that shows "no binding at pH 6.0" may just be a binder that doesn't survive pH 6.0. Adaptyv x Anthropic Challenge 4 is open: design a TfR1 binder that holds at pH 7.4 and lets go at pH 6.0, for delivery across the blood-brain barrier.