Iktos is the only company offering state-of-the-art, ready-to-use generative modeling solutions with built-in synthetic accessibility for successful drug discovery projects.
At Iktos, we have developed an innovative technology platform for drug design and discovery. Generative AI is taking off and a myriad of deep generative models have been reported in the scientific literature. Incorporated in 2016, Iktos has pioneered Generative AI work and stands at the forefront of AI-based generative chemistry by combining its internal R&D effort with a continuous benchmarking effort to incorporate new methods in our technology platform. Our technology has delivered value in 50 plus real-world research collaborations with some of the most eminent pharma companies globally.
Generative models relying on deep neural networks have recently emerged as promising tools to explore the chemical space in early or late-stage discovery programs, generate valuable ideas for chemists and overcome multi-parametric optimization (MPO) challenges, by proposing AI-designed optimal molecules that match your target product profile (TPP) in a relatively short time frame.
Complementary to your design efforts, Iktos generative AI technology provides new, unbiased insights on the project, based on an optimal exploitation of the information hidden in your data. From hit discovery to lead optimization, our technology integrates ligand and structure-based approaches to design new molecules solving your project objectives. Our proprietary algorithms excel at multi-parametric optimization (MPO).
- If you are struggling to make the consensus on all the objectives of a lead optimization project (10 and more), Iktos will help you design new structures, close to your chemical series, that optimize all your endpoints.
- If you are looking for new ideas, new cores or scaffolds, Makya will complement your creativity thanks to its unique generative AI trained to mimic chemist’s intuition
- If you want to discover new hits based on 2D, 3D signature similarity (e.g. pharmacophores, shape or charges) or crystallography data, Iktos is developing proprietary methods to design virtual hits with new scaffolds, for any given blueprint.
We have developed Makya, a fully automated SaaS platform for de novo drug design that enables rapid identification of molecules that satisfy multiple drug-like criteria to expedite drug discovery and development. Makya is suitable for both ligand-based and structure-based virtual lead design and allows multi-parametric optimization (MPO) of lead molecules towards potential pre-clinical drug candidates
Multiple generative methods covering different use cases
Makya has three main generative AI algorithms which cover most of the use cases in drug design:
Fine tuning generator:
Generation around a chemical space (dataset, single molecule, chemists’ ideas…)
- Technology: AI based on the SMILES representation using LSTM architecture powered by reinforcement learning
- Low diversity
- Goal: Find an optimal solution within your chemical space in line with the constraints of your project
Generation around a single reference molecule
- Technology: AI based on the SMILES representation using a Transformer architecture trained on patent literature data with the goal to mimic chemists’ intuition to create novel and diverse molecular designs relevant to your project
- Medium to High diversity
- Goal: Find new ideas with high novelty, useful for re-scaffolding / new hit discovery
Generation starting from an advanced synthesis intermediate
- Technology: Based on Iktos synthetic planning technology (Spaya), the Forward generator is able to generate molecules by combining your intermediate and commercial compounds following your set of constraints (allowed reaction site and reaction types).
- Low to High diversity
- Goal: Creation of a focused library, easily accessible from commercial starting materials
Our generative design platform also integrates key supporting technologies developed by Iktos which are key to guarantee the quality of our results in all your use cases:
We have introduced fitness functions to guide the molecular generators in exploring the vast chemical space and to produce optimally designed molecules in line with project needs. The fitness functions are predictors, specifically built from the collaborator’s project’s data set: molecules, measured bioactivity data points, activity, selectivity, ADMET… but also based on public and private datasets. In addition, we also have implemented fitness functions to maximize druggability, synthetic accessibility (spaya.ai), and similarity to collaborators initial data set. The molecules generated from Iktos technology platform can be optimized simultaneously on multiple parameters in line with the success criteria.
Top Notch Predictive Models
Iktos molecule generation technology is only as good as the predictors it uses to travel in the chemical space. Therefore, the quality of the predictors is key. We have developed an auto machine learning (auto-ML) module which automatically and in a very short time frame, constructs and selects high-performance predictors based on your data, by benchmarking and optimizing both machine learning algorithms and molecular descriptors (1D, 2D, 3D or data-driven molecular representations), to achieve optimal predictive performance metrics.
Built-in Synthetic Accessibility Prediction
Generative models are known for their creativity and can sometimes end up with crazy and unfeasible molecules. Iktos has solved this problem by integrating its AI for synthetic planning technology, initially built for Spaya, into our generative design pipeline. At each step of the generative process, the molecule generator is rewarded when it designs synthetically accessible molecules and punished if it comes up with crazy structures. This key feature guarantees an excellent level of quality and feasibility of the generated molecules, and furthermore can be customized to your requirements (building blocks, reaction space, number of steps…)
Structure-Based Generative Design
Generative modeling when applied in combination with structure-based design, allows users to quickly and efficiently navigate through the chemical space, to find relevant 3D solutions, balanced physicochemical properties and FTO of the proposed candidates. This approach allows convergence to an optimal number of molecules during the generation and 3D validation. Multiple criteria can be factored in the compound generation (such as optimal physchem characteristics) alongside biological activity which will increase the chance of success of future optimization.
Spaya is Iktos’s computer-aided synthetic planning technology that uses data-derived AI models to discover and prioritize synthetic routes. The underlying AI engine has been trained on several millions of reactions and Iktos continues to train it with more reaction data, continuously enhancing its performance.
Spaya performs an exhaustive analysis of all possible synthetic routes towards a given compound and ranks them according to various criteria, each route ending with commercially available building blocks.
Iktos synthetic planning technology is fully data-driven and can be customized to your needs by adding your reaction data, your proprietary building blocks and preferred providers. The technology is available within Spaya GUI designed for synthetic chemists with many advanced features available, and can also be used through a dedicated API (Spaya API) enabling to process a large number of compounds in a minimal time, thanks to Iktos scalable cloud infrastructure. Iktos synthetic planning technology is also natively included in our generative design platform.
Expertise and Capabilities
At Iktos, we have a brilliant and talented team with unique set of capabilities in the field of AI and drug discovery. Iktos’s mission is to provide AI technology for medicinal chemistry to accelerate new drug discovery, increase success rates and improve productivity. The Iktos team has a comprehensive skills set across data sciences, computer sciences, ML and AI; molecular modelling, cheminformatics; medicinal chemistry, drug discovery; IT and data engineering with experience in HPC and cloud infrastructure; business development and marketing disciplines. Iktos team of PhDs and Engineers are well equipped with the skills and resources needed to successfully deliver a drug discovery project, deliver solutions that fit your user, technical and security requirements, and support you in the deployment, operation and use of our software.
A State-of-the-art HPC Infrastructrure
Iktos has developed its own HPC platform that enables to parallelize computation on GPU and CPU on demand, and can easily be scaled up to meet any computation needs. All Iktos technologies can also easily be deployed on your virtual private cloud (VPC) either on AWS, Azure or Google if needed.