We start by meeting to review your project and what data you have already generated. We will determine what stage you’re at and what additional steps are necessary to get you where you want to go.

We will perform a careful review of the academic and patent literature around your target, and curate and add relevant data from the literature to increase the size of our dataset. This can improve the predictive ability of models we build. Any data we discover will be discussed with you to ensure its validity and appropriateness for incorporation into models. We will carefully standardize all data by neutralizing compounds, finding preferred tautomers, removing metal ions and salts, and removing duplicates.

We will then use our AI algorithms to construct QSAR models based on your data, that which we find from literature, and our novel molecular/atomic descriptors. Conversely, we can incorporate any QSAR/QSPR models that you may have generated internally.

Prior to performing the first AIDD run, we’ll put on our medicinal chemist hats and closely inspect the data to better guide the AIDD process. This includes utilizing best-in-class cheminformatic toolkits, identifying key activity cliffs via matched molecular pair analysis (MMPA), generating classes and scaffolds for each class, then generating and analyzing R-tables to provide clues for which parts of the molecule may be the best to evolve.

We will then perform our first AIDD run with parameters for optimization decided upon collaboratively.

We typically optimize 4-5 parameters simultaneously, the most common parameters being target activity, ADMET Risk™, preclinical/clinical PK endpoints (e.g., bioavailability) and synthetic difficulty. However, there are over 100 parameters for which your molecules can be optimized.

Our experts will then filter results for novelty (by similarity searches against patent databases) and commercial availability (by searching chemical manufacturer databases such as Enamine REAL). If these searches limit the potential number of molecules, a second round of AIDD will be performed.

Our experts will begin the collaborative research project by meeting with your team to discuss the overall goals of the project, the HTS assay design, makeup of the initial screening library, and any initial screening results should you already have them. We can then help you design an initial screening library or follow-on/focused libraries and/or analyze the hits you have to select the most promising leads for further development.

We understand the quality of your screening library significantly impacts your results. It is important to consider not only the chemical space appropriate for your target, but also the physiochemical properties of the compounds themselves. Successful translation from hit to lead requires that the compound have appropriate ADMET and PK properties. We can use our industry-leading ADMET Predictor platform to predict important ADMET and PK properties and limit library constituents to only those meeting defined thresholds, thus increasing the likelihood of hits that are lead-like/drug-like. We can also filter results by commercial availability, whether it be fully enumerated compounds or building blocks for combinatorial chemistry approaches.

We can also help design follow-on or focused library builds after a diverse compound library HTS. Using data derived from the initial hit population, we can apply our combinatorial chemistry library enumeration tool in the MedChem Studio™ Module to create virtual libraries and work with you to facilitate purchase of compounds to build that library.

Analyzing the results of a HTS campaign is also a crucial step in the lead discovery process. Our team has tools to streamline HTS hit analysis and maximize the chance of downstream success. As a first step, we can quickly and accurately predict more than 100 important PK and ADMET properties for every hit, including our unrivalled pKa prediction, bioavailability, as well as synthetic difficulty. We also provide easy-to-comprehend risk plots for ADMET Risk™, toxicity risk, CYP inhibition risk, and more. We can generate classes containing the same scaffold, predict those same classes, and compare relative activity and risk among the classes. This can help with the selection of a lead scaffold or scaffolds for further optimization. Furthermore, MMPA can be performed to identify important activity cliffs to guide future evolution of the lead molecule(s).

Let us help your team decide the best candidates for further testing and optimization. Data from HTS campaigns can also be used to build QSAR models using ADMET Modeler™, which can help inform additional library builds and medicinal chemistry optimization efforts.

A typical project begins by meeting with your team to discuss the overall goals of the project and data you have generated that could be used for model building. We can also perform a careful review of the academic and patent literature around your target, and curate and add relevant data from the literature to increase the size of our dataset and improve the predictive ability of models we build. Any data we discover will be discussed with you to ensure its validity and appropriateness for incorporation into models. We will also work with you to decide how the data should be divided for training and testing, and what method(s) are most appropriate for parsing of the data.

After collaboratively curating the dataset, we will carefully standardize all data by neutralizing compounds, finding preferred tautomers, removing metal ions and salts, and removing duplicates. Next, we will use our custom AI/ML algorithms to serially develop predictive models, using results from one round of builds to inform subsequent builds. We have a variety of techniques to generate both classification and regression models to ensure the best possible models are created.

We can also predict a spectrum of ADMET and PK properties with ADMET Predictor®, including pKa(s), solubility vs. pH, logD vs. pH, CYP & UGT metabolism/inhibition, Ames mutagenicity, skin and respiratory sensitivity, rat and mouse TD50, and much more. Descriptor and structure sensitivity analysis (DSA and SSA) tools can then be used to determine what atomic and molecular parameters contribute most to the QSAR/QSPR models and which atoms in individual structures contribute most to the model score, which can be used to guide future lead optimization efforts.