
join us for an evening of light networking and drinks!
RL unlocked AI for coding + math can it do the same for science? we're hosting a casual drinks+networking session to discuss challenges and oppos on the 27th of August Luma link in the comments!

AI + data + biology

join us for an evening of light networking and drinks!
RL unlocked AI for coding + math can it do the same for science? we're hosting a casual drinks+networking session to discuss challenges and oppos on the 27th of August Luma link in the comments!

1/ our mission at @SciMac has always been to accelerate the discovery of new medicines - and life science more broadly - using AI. but we never really touched on why acceleration is so important, beyond the obvious

Watch our launch video and explore the new ScienceMachine -> sciencemachine.ai

We will publish more results soon. This marks an inflection point for ScienceMachine and for the wider industry, as AI starts to unblock the regulated, slow, and process-heavy routine that comes with real-world drug development.

Our team has been working hard to reach new levels of AI accuracy and reliability in this field, enabling the ingestion of terabytes of data, and the automation of the most manual and tedious tasks, such as raw data processing, review of entire studies, and regulatory reporting.

We are building the world's first AI automation platform purpose-built for regulated drug development: reliable, compliant, secure and traceable - so it can actually be used in high-stakes environments.

Today, at ScienceMachine we're sharing more about the groundbreaking work we’ve been doing with leading companies around the world. More in the thread 🧵

🤖🧬
A lot of scientific work is slower than it should be for a surprisingly simple reason: powerful tools still need people to move the work between them. In bioanalysis, teams already have tools they trust. SCIEX OS, Skyline, Analyst, Thermo, Waters, internal scripts. They are often very good at the specific jobs they were built for. The problem becomes obvious when you zoom out from any one tool and look at the whole workflow: a file gets moved from one place to another. A folder has to be prepared in the right format. A method has to be selected. A run has to be configured. Outputs need to be checked, copied, reviewed, and turned into something useful for the next step. None of this is exactly glamorous work. But it has to be done and done carefully. And when you watch highly trained scientists spend hours on it, you start to feel how much human attention is being spent on work that is necessary, but not very creative. We recently built an integration between ScienceMachine and SCIEX OS for one of our customers. The idea was simple: do not replace the software the team already trusts. Instead, make it part of a larger workflow that ScienceMachine can help orchestrate. Read the full technical write-up in the comments 📷 ScienceMachine can now reach into a real customer environment, work with SCIEX OS in the place it already runs, and turn a manual desktop workflow into a secure, repeatable step in a larger scientific process. The best part is that every integration like this compounds. Once ScienceMachine can work with one important tool in the workflow, it becomes easier to connect the next one, and then the next one, until more of the scientific process starts to feel continuous instead of fragmented.

A lot of scientific work is slower than it should be for a surprisingly simple reason: powerful tools still need people to move the work between them. In bioanalysis, teams already have tools they trust. SCIEX OS, Skyline, Analyst, Thermo, Waters, internal scripts. They are often very good at the specific jobs they were built for. The problem becomes obvious when you zoom out from any one tool and look at the whole workflow: a file gets moved from one place to another. A folder has to be prepared in the right format. A method has to be selected. A run has to be configured. Outputs need to be checked, copied, reviewed, and turned into something useful for the next step. None of this is exactly glamorous work. But it has to be done and done carefully. And when you watch highly trained scientists spend hours on it, you start to feel how much human attention is being spent on work that is necessary, but not very creative. We recently built an integration between ScienceMachine and SCIEX OS for one of our customers. The idea was simple: do not replace the software the team already trusts. Instead, make it part of a larger workflow that ScienceMachine can help orchestrate. Read the full technical write-up in the comments 📷 ScienceMachine can now reach into a real customer environment, work with SCIEX OS in the place it already runs, and turn a manual desktop workflow into a secure, repeatable step in a larger scientific process. The best part is that every integration like this compounds. Once ScienceMachine can work with one important tool in the workflow, it becomes easier to connect the next one, and then the next one, until more of the scientific process starts to feel continuous instead of fragmented.

Big thanks to the Inoviv team for the collaboration so far. Excited for what’s ahead! Full case study here: drive.google.com/file/d/1HWw…

Excited to announce ScienceMachine’s partnership with Inoviv to bring AI automation into proteomics workflows. 🚀 Together, we’ve reduced time spent on QC and reporting by ~80% and increased lab capacity by 20% — giving scientists more time to focus on science, not admin.

How capable are AI agents at bioinformatics research? I tested 9 LLMs on BixBench: 205 complex tasks from real-world research – the kind I do as a compbio PhD Results: - 63% accuracy (3x higher than the SOTA 10 months ago) - The harness mattered as much as the model

So we are asking for your advice: try out our platform with your most tedious scientific tasks and let's schedule a call on how we can improve the platform!

But with her and others’ expert feedback, we went heads-down building and shipped a fine-tuned product, fit-for-purpose for complex scientific work. And we’re only getting started!

But it was not always like this! She first tried our platform a few months ago and left it feeling… underwhelmed. It just didn’t understand what she needed!

Marija is a PhD student at King's College London, pushing the boundaries of our understanding of RDEB -- a severe genetic skin disorder causing extreme skin fragility, blistering, and wounds. For her, ScienceMachine makes the frantic sprint for her thesis deadline into a breeze