MolBioMike (@MolBioMike) header
MolBioMike (@MolBioMike) profile picture

Biochemist with a DIY attitude. I experiment, optimize, and share ways to make bench science cheaper, clearer, and more effective.

Colorado, USA
Joined January 2026
Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)Photo by MolBioMike (@MolBioMike)
Lodgepole is built different 💪🏻 We are three PhD scientists from Regeneron Ginko and Plasmidsaurus. We have built the therapeutics, the tools, and the assays to understand them. I have been head of chemistry at a msmAb ADC startup and understand what a target engagement study and efficacy study means and how to design one. I have seen promising candidates die in CMC. I know what a mab should do and should not do in a vial upside down for six months in a fridge. Let us run your accelerated stability studies or build your lot release ELISA. Did I mention I have more than a decade of wet lab neuro experience in academia and industry? I can work closely with your scientific team on your amyloid or trancytosis projects. Speaking of industry, I have been a part of dozens of IND committees and helped push multiple drugs over the filing finish line. I have seen drugs fail and can help your team get to a go/no-go faster and with more clarity. We are not a couple of college dropouts with millions of dollars of VC. I actually built a real therapeutics lab. With my money that I earned from selling my time and expertise. Sure you can send us work via a Cloudflare tunnel and API. Yes we can work with your team to give access to our models. Yes we can use more opentrons and robot arms. But more importantly we actually know how to build molecules, test molecules, and report results. Soon we will have a flashy set of posts bragging about our models and our badass wet lab but underneath it is real substance and expertise.
Why would someone choose this CRO over another? They have no experience actually doing experiments and no way the cost is less. People are supposed to pay $5000 or $10,000 for you to run an experiment for the first time in your life and we are supposed to trust the results. It’s not faster and cheaper if the experiment wasn’t done correctly.
5
2
39
2,653
MolBioMike retweeted
Listening to MFers with NO pipette swag talking about lab automation
5
15
2
117
3,490
Replying to @biotech_hunter
Why would I want to downsize
1
1
6
115
MolBioMike retweeted
These guys are new and definitely worth talking to! It's a small team with a ton of expertise and experience.
We are live and accepting new customers. Pharma grade Binder design -> Lab testing The team includes ex Regeneron, ex Ginko, and ex Sartorius scientists. We are protein biochemists and cell biologists on a mission. We move fast. Let’s get on a call.
2
3
14
1,707
MolBioMike retweeted
Replying to @hhlee
Helix in the streets and a barrel in the sheets, get you one that can do both
2
1
12
531
We are live and accepting new customers. Pharma grade Binder design -> Lab testing The team includes ex Regeneron, ex Ginko, and ex Sartorius scientists. We are protein biochemists and cell biologists on a mission. We move fast. Let’s get on a call.
I’ve spent >>$100k of my own money building this. End to end lab capabilities coming online to support discovery research. New lab space with significantly more bench space. Molecular biology: cloning, libraries, subcloning, mutagenesis Protein expression: E. coli, CHO, HEK, etc Protein purification: all common resin chemistries (affinity, IEX, SEC, custom columns) Protein QC: gels, two color WB, ELISA, SPR, mass photonics, TEM In vitro assays: effector function, internalization, efficacy, dose response In vivo: pk and pharmacology (at our partner labs) Industry standard equipment from top brands including Thermo, Cytiva, Waters, and LiCor.
1
3
1
14
2,230
So I’m talking to a new prospective client.. running through the typical questions.. Who are you? What are you looking for? How did you find me? “I was asking Astra who are the best protein engineering consultants and it said you.” So I asked for a poster from Astra LFG
1
1
8
455
Replying to @mgdurrant
Well, we would if we didn’t have biosafety triggers popping up with the slightest hint of biology happening in the prompt.
1
1
155
Finally a more intuitive way to learn pLDDT/pAE? 😎 sokrypton.github.io/protein_… (Character idea from @HannesStaerk & Alex Waldherr)
56
287
44
1,403
166,403
20k folds in an hour on CPU. Try jaxfvld today!
5
4
1
40
3,395
Patent pending status.
Big milestone for Fabricagen: we’ve officially filed our first patent application and secured our priority date. 🤠
1
4
406
We just released our second open-source project in a week jaxfvld is a 760k-parameter, fully differentiable antibody structure predictor built in JAX On a leak-free benchmark it essentially ties IgFold on CDR-H3 accuracy (and other metrics) It runs on a commodity CPU, folds an Fv in under a second per seed, and the code and weights are open. This is the kind of tool I wanted to exist, so I built it, and I'm shipping it to you I hope you find it as useful for your workflows as it already has been for us This model is useful for folding-in-the-loop and other advanced applications in protein ML I would love to hear your thoughts. DMs open.
Fabricagen is excited to announce the release of our new folding model, jaxfvld. jaxfvld is a small, fast, fully differentiable antibody Fv structure predictor written in JAX/Flax. It folds an antibody variable region (paired VH and VL, or a single VHH domain) from sequence to an all-atom backbone, returning: per-residue frames, side-chain torsions, atom14 coordinates, and a per-residue pLDDT. It takes no antigen sequence or structure and no input structure (though structures can be used to guide folding).
5
4
27
2,184
Here is a different way to look at the data. Three proteins optimized and their iPTMs were unaffected.
1
1
1
1
827
It’s out!
Our JAX Fv folder was observed to have near CDRH3 structural recovery parity with IgFold using our internal leak free corpus We benchmarked against our internal hold outs and their original set rederived 0.1 A different right now 2.908 vs 2.812 I am going to push it further with a different corpus xtal:synth sampling ratio and push for more steps Either way we will share this tool and the sister jax tool we built, ablanx, this week. Weight code benchmarking data and permissive licensing. Jaxfvld is lightweight and is sequence differentiable, allowing it to be used in interesting applications. Ablanx is a faithful jax implementation of Ablang2 useful for differentiable antibody workflows. We needed it so we could recreate an IgFold like repo differentiable to sequence. So we built it. And we are shipping it to you. We will drop a technical document on ablanx tomorrow with code and weights. Jaxfvld will follow after some final benchmarking and testing. Lastly, a developability focused composition of the two, called sift, will be available for testing via a web portal as soon as I can finish everything. Once the code and weights are up please test it hard and let me know what you think!
1
1
4
930
Replying to @FabricagenDOTai
@FabricagenDOTai released jaxfvld today and we want you to put it through its paces. Let us know if you have any questions or comments!
I spent some time building a tool I wish I had before A lightweight, MSA free, leak free sequence differentiable fv folder The last bit of training and benchmarking is happening now using a proven recipe with more steps The write up and repos are ready to ship but i wanted to take one more shot on goal. Looking forward to the reviews from the community
1
4
441
jaxfvld is released under BSD-3-Clause. The repository includes: – model code and weights – benchmark fixtures – parity and calibration reports – reproducible scoring commands – paired Fv and VHH inference – integration with the pure-JAX AbLang2 port, ablanx pip install jaxfvld This is not intended to replace the largest structure models Jax or otherwise. It is meant to be a compact, hackable, differentiable antibody folder built for researchers who want structure prediction to become part of the design loop. Code: github.com/fabricagen/jaxfvl… We would love to hear what you think of this project.
2
5
172
MolBioMike retweeted
Thank you to everyone who spent their Saturday with us at Ebolathon—and congratulations to the BuilderBlockers, Michael, @cydatio, and Rafal, who took an interesting synthon-aware Thompson Sampling approach to guide compound selection. All submissions have been sent to our co-host, @onepot_ai, for synthesis before moving into pseudovirus assay testing. All assay data will be made public. A special thank you to our judges—@andrei_tyrin, Andrii Kyrylchuk, Amy He, and Nina Jovic—for lending their time and expertise. And thank you to our sponsors—@RowanSci, @boltz_bio, @OpenRouter, @modal, and @anyscalecompute for giving participants access to the models and compute needed to explore the challenge, and to @posthog for hosting us in their wonderful space!
3
8
2
26
2,525
A sneak peek of our new folder #JaxFvld coming to a repo near you We have built a smaller model and achieved parity with #IgFold This lightweight Jax Fv folder is part of our suite of Jax tools allowing lightweight folding in the loop without heavy AF computation We used an expanded corpus (thx SabDab) and used a very similar training recipe as the original work. Looking forward to sharing these benchmarks with you in a couple days
1
13
1,274
We just open-sourced ablanx. A faithful JAX/Flax implementation of the AbLang2 antibody language model that loads the original published weights unchanged. Features: • pip install ablanx • Numerically validated against the PyTorch reference • JIT / vmap compatible • BSD-3 licensed This is the first building block in our open antibody ML ecosystem. GitHub: github.com/fabricagen/ablanx Next up: jaxfvld (differentiable antibody folding) and sift (paratope-aware developability optimization).
2
1
19
3,182