Projects
Our Innovative Work
At Synthexo, our portfolio highlights real-world applications of our AI tools in the biosciences
Protein Design
Our AI-driven approach transforms protein engineering through advanced algorithms and Design-Build-Test-Learn cycles.
Strain Engineering
Revolutionizing the strain engineering process through AI and Computational Biology, we help create high-yielding host strains and vectors.
Collaboration
We integrate data and models from frontier research labs and partner with leading institutions to foster innovation through shared research.
AI Architectures
Explore our diverse AI architectures designed for tackling complex challenges. We optimise for CPU / GPU / TPU as required, given the learning task and available compute platform.
AI Training and Optimisation
Huge energy and cost savings could be obtained when architectures are optimised. Explore how we could optimise your training and architectures to save money and lower your carbon footprint
Discovery
Data-driven discovery in complex datasets is part of our DNA. Hypothesis-free discovery of deep patterns in NGS and other omics has changed the way discovery is done. We employ state-of-the-art AI and statistical enrichment algorithms, combined with rich visualisation to bring data to life.
Frequently Asked Questions
Your Questions Answered
Is AI a large language model?
The transformer architecture that has become synonymous with AI is indeed a very generic architecture and has been behind some important breakthroughs in natural language processing, protein modelling and many other scientific and non-scientific applications. However, the transformer architecture (and its many flavours and attention methods) is only a small part of the infrastructure associated with the training and inference of AI models. Other AI models and methods that Synthexo is also employing include, but are not limited to, are large quantitative models, convolutional neural networks, multilayer perceptrons, mixture-of-experts, retrieval augmented generation, search, random forests, linear models, likelihood networks, reinforcement learning, diffusion, flow matching, Bayesian inference and active inference. Numerous design decisions are also to be considered, such as embeddings, optimal vector search, databases, memory constraints, bottlenecks, and others.
Do I need to train a model from scratch?
Often, but not always, the best accuracy can be gained by training a model from scratch, given sufficient data, capturing the problem you are modelling. However, in many cases, you may not have enough resources to gather the correct amount and diversity of data. This is were foundational open source models are very useful. A good example is protein large language models. They have been trained on vast sets of protein sequences, in a process called self-supervised learning. using them zero-shot (off-the-shelf, as-is) could already give you in some cases, a useful output. However, it is when these models are used as embeddings in a transfer learning (few-shot, downstream tasks) that we see them as an effective tool for capturing a wealth of information. We offer tailored solutions for using the best-in-class foundational models in bespoke algorithms and workflows for your task, in which we could re-train on a more affordable custom subset of data. Effectively, this translates the wealth of information gained on well-funded global academic projects, into a language that leads to actionable insights for you bio project.
What Industries can Benefit from AI?
Industries such as pharmaceuticals, biotech, healthcare and agriculture can benefit immensely from AI applications, driving innovation and improving outcomes. More so, it enriches the human learning experience in that it helps reveal deep patterns in the data that the human eye cannot detect easily, and make the work that we do in the lab, more productive. More successful results means you can do fewer iterations, so that you can continue doing what you love to do, whether it is to be working with a pipette, doing data analyses, or working in the field.
How to Get Started?
Getting started is easy. Contact us for a consultation, and we will guide you through the process of integrating our tools or models into your research.