At Raven biosciences, we harness the power of cutting-edge technologies, including AI and biophysical models, to revolutionize molecule design and accelerate development across all major modalities. Our innovative approach ensures precision and efficiency in creating small molecules, macrocycles, antibodies, peptides, enzymes, DNA, and RNA.
Why Raven biosciences?
Computer-assisted drug discovery is an integral part of identifying hits and developing leads in the modern pharmaceutical industry.
Our in silico portfolio cover high-throughput virtual screening, hit-to-lead-, and lead optimization regarding target affinity, target selectivity, and ADMET properties using docking, molecular dynamics, and QSAR/QSPR technologies.
For unknown target protein structures, our experts build homology models built on a combination of related proteins and experimental data.
We have also developed an in-house platform for ligand-based drug discovery – ChemX. With ChemX you can train models to predict toxicity, binding affinity and potency to name a few.
The methods we apply to molecule development can be applied in many other applications, e.g. improving chemical probes for in vitro assays.
Macrocycles and peptides are highly useful modalities in modern pharmacology, but their size hinders the use of many computational methods developed for small molecules.
None-the-less, macrocycles and peptides are excellent choices of modality if the targeted binding site more closely resembles a shallow pool, or if the site is a protein/protein interface. Additionally, their safety profiles differ from typical small molecules, making them desirable modalities for many diseases.
We have therefore built custom workflows for these modalities, making virtual screening and lead optimization possible. Regardless of the desired macrocycle being peptide-based, synthetic-polymer based, or something else entirely.
At Raven biosciences, we span both traditional bioinformatic technologies and new DNA and RNA language models.
As an example, we perform in-depth analysis of RNA-Seq data to compare sequencing data generated using different protocols. Development of code and integration of existing tools to analyse gene coverage, intron retention, fusion genes, internal priming, and secondary structure features.
We also apply AI to predict downstream properties of DNA primers and functional RNAs, such as efficacy, function, and stability. These predictions are complemented by our computational biophysics expertise in 3D modeling, including mapping protein–RNA interactions to provide structural context and prioritize designs.
Antibody modeling is notoriously challenging, and remain an area of difficulty even in the era of AI. We have developed our own platform for characterization and design – EpiC.
Combining extensive physics-based modeling with clever clustering and affinity prediction, we can determine the correct interface 85% of the time.
EpiC has determined the antibody/antigen interface in cases where experimental methods fall short due to highly dynamic antigens or epitopes consisting entirely of loop regions. EpiC has also been adapted to handle nanobodies.
Recent breakthroughs in AI-driven protein structure prediction and generative sequence design have transformed in-silico enzyme engineering from niche research into mainstream industrial practice.
Whether your goal is improving an existing enzyme candidate or expanding your IP through metagenomic screening of ortholog enzymes, we can help achieve it.
Our consultants utilize state-of-the-art AI models and physics-guided scoring to pre-select mutations enhancing enzyme properties such as thermostability, solubility, activity, and substrate specificity.
While new models have dramatically enhanced protein design capabilities, we also highlight the necessity of leveraging enzymes already optimized by nature. Thus, we offer targeted metagenomic screening of ortholog enzymes based on both structural and sequence alignment, significantly enriching and securing your IP portfolio.
We have solved many projects that fall outside the typical use cases of the methods outlined above and we love a challenge! We have therefore listed two examples of atypical modelling, and hope you.
In the first case, a group of scientists found a protective mutant of a protein relevant to a degenerative disease. They wanted to understand how the mutant affected the behavior of the protein and its normal function, and evaluate the likelihood of designing a small molecule therapeutic that could mimic the effect. We modeled the protein in multiple conformations relevant to several biological pathways, and assessed behavior using molecular dynamics simulations of both wild-type and mutant protein.
In the second case, a therapeutic company has developed a series of compounds with a desired biological effect, but they have a hard time identifying the target protein. Each experimental approach they applied returned a list of targets with no common overlap. We modeled the binding of the compound series to the proposed targets and compared the ligand-based structure/potency relationship as an evaluation metric to identify the target protein.
