Comparison of logP and logD correction models trained with public and proprietary data sets.

In drug discovery, partition and distribution coefficients, logP and logD for octanol/water, are widely used as metrics of the lipophilicity of molecules, which in turn have a strong influence on the bioactivity and bioavailability of potential drugs. There are a variety of established methods, mostly fragment or atom-based, to calculate logP while logD prediction generally relies on calculated logP and pKa for the estimation of neutral and ionized populations at a given pH. Algorithms such as ClogP have limitations generally leading to systematic errors for chemically related molecules while pKa estimation is generally more difficult due to the interplay of electronic, inductive and conjugation effects for ionizable moieties. We propose an integrated machine learning QSAR modeling approach to predict logD by training the model with experimental data while using ClogP and pKa predicted by commercial software as model descriptors. By optimizing the loss function for the ClogD calculated by the software, we build a correction model that incorporates both descriptors from the software and available experimental logD data. Additionally, we calculate logP from the logD model using the software predicted pKa's. Here, we have trained models using publicly or commercial available logD data to show that this approach can improve on commercial software predictions of lipophilicity. When applied to other logD data sets, this approach extends the domain of applicability of logD and logP predictions over commercial software. Performance of these models favorably compare with models built with a larger set of proprietary logD data.

A probabilistic method to report predictions from a human liver microsomes stability QSAR model: a practical tool for drug discovery.

Using data from the in vitro liver microsomes metabolic stability assay, we have developed QSAR models to predict in vitro human clearance. Models were trained using in house high-throughput assay data reported as the predicted human hepatic clearance by liver microsomes or pCLh. Machine learning regression methods were used to generate the models. Model output for a given molecule was reported as its probability of being metabolically stable, thus allowing for synthesis prioritization based on this prediction. Use of probability, instead of the regression value or categories, has been found to be an efficient way for both reporting and assessing predictions. Model performance is evaluated using prospective validation. These models have been integrated into a number of desktop tools, and the models are routinely used to prioritize the synthesis of compounds. We discuss two therapeutic projects at Genentech that exemplify the benefits of a probabilistic approach in applying the models. A three-year retrospective analysis of measured liver microsomes stability data on all registered compounds at Genentech reveals that the use of these models has resulted in an improved metabolic stability profile of synthesized compounds.

An integrated suite of modeling tools that empower scientists in structure- and property-based drug design.

Structure- and property-based drug design is an integral part of modern drug discovery, enabling the design of compounds aimed at improving potency and selectivity. However, building molecules using desktop modeling tools can easily lead to poor designs that appear to form many favorable interactions with the protein's active site. Although a proposed molecule looks good on screen and appears to fit into the protein site X-ray crystal structure or pharmacophore model, doing so might require a high-energy small molecule conformation, which would likely be inactive. To help scientists make better design decisions, we have built integrated, easy-to-use, interactive software tools to perform docking experiments, de novo design, shape and pharmacophore based database searches, small molecule conformational analysis and molecular property calculations. Using a combination of these tools helps scientists in assessing the likelihood that a designed molecule will be active and have desirable drug metabolism and pharmacokinetic properties. Small molecule discovery success requires project teams to rapidly design and synthesize potent molecules with good ADME properties. Empowering scientists to evaluate ideas quickly and make better design decisions with easy-to-access and easy-to-understand software on their desktop is now a key part of our discovery process.

Identification of Macrocyclic Peptide Families from Combinatorial Libraries Containing Noncanonical Amino Acids Using Cheminformatics and Bioinformatics Inspired Clustering.

In the past decade, macrocyclic peptides gained increasing interest as a new therapeutic modality to tackle intracellular and extracellular therapeutic targets that had been previously classified as "undruggable". Several technological advances have made discovering macrocyclic peptides against these targets possible: 1) the inclusion of noncanonical amino acids (NCAAs) into mRNA display, 2) increased availability of next generation sequencing (NGS), and 3) improvements in rapid peptide synthesis platforms. This type of directed-evolution based screening can produce large numbers of potential hit sequences given that DNA sequencing is the functional output of this platform. The current standard for selecting hit peptides from these selections for downstream follow-up relies on the frequency counting and sorting of unique peptide sequences which can result in the generation of false negatives due to technical reasons including low translation efficiency or other experimental factors. To overcome our inability to detect weakly enriched peptide sequences among our large data sets, we wanted to develop a clustering method that would enable the identification of peptide families. Unfortunately, utilizing traditional clustering algorithms, such as ClustalW, is not possible for this technology due to the incorporation of NCAAs in these libraries. Therefore, we developed a new atomistic clustering method with a Pairwise Aligned Peptide (PAP) chemical similarity metric to perform sequence alignments and identify macrocyclic peptide families. With this method, low enriched peptides, including isolated sequences (singletons), can now be clustered into families providing a comprehensive analysis of NGS data resulting from macrocycle discovery selections. Additionally, upon identification of a hit peptide with the desired activity, this clustering algorithm can be used to identify derivatives from the initial data set for structure-activity relationship (SAR) analysis without requiring additional selection experiments.

Handling of tautomerism and stereochemistry in compound registration.

Automated registration of compounds from external sources is necessitated by the numerous compound acquisitions from vendors and by the increasing number of collaborations with external partners. A prerequisite for automating compound registration is a robust module for determining the structural novelty of the input structures. Any such tool needs to be able to take uncertainty about stereochemistry into account and to identify tautomeric forms of the same compound. It also needs to validate structures for potential mistakes in connectivity and stereochemistry. Genentech has implemented a Structure Normalization Module based on toolkits offered by OpenEye Scientific Software. The module is incorporated in a graphical application for single compound registration and in scripts for bulk registration. It is also used for checking compounds submitted by our collaborators via partner-specific Internet sites. The Genentech Structure Normalization Module employs the widely used V2000 molfile format to accommodate structures received from a wide variety of sources. To determine how much information is known about the stereochemistry of each compound, the module requires a separate stereochemical assignment. A structural uniqueness check is performed by comparing the canonical SMILES of a standard tautomer. This paper offers a discussion of the steps taken to validate the chemical structure and generate the canonical SMILES of the standard tautomer. It also describes the integration of the validation module in compound registration pathways.

DEGAS: sharing and tracking target compound ideas with external collaborators.

To minimize the risk of failure in clinical trials, drug discovery teams must propose active and selective clinical candidates with good physicochemical properties. An additional challenge is that today drug discovery is often conducted by teams at different geographical locations. To improve the collaborative decision making on which compounds to synthesize, we have implemented DEGAS, an application which enables scientists from Genentech and from collaborating external partners to instantly access the same data. DEGAS was implemented to ensure that only the best target compounds are made and that they are made without duplicate effort. Physicochemical properties and DMPK model predictions are computed for each compound to allow the team to make informed decisions when prioritizing. The synthesis progress can be easily tracked. While developing DEGAS, ease of use was a particular goal in order to minimize the difficulty of training and supporting remote users.

chemalot and chemalot_knime: Command line programs as workflow tools for drug discovery.

BACKGROUND: Analyzing files containing chemical information is at the core of cheminformatics. Each analysis may require a unique workflow. This paper describes the chemalot and chemalot_knime open source packages. Chemalot is a set of command line programs with a wide range of functionalities for cheminformatics. The chemalot_knime package allows command line programs that read and write SD files from stdin and to stdout to be wrapped into KNIME nodes. The combination of chemalot and chemalot_knime not only facilitates the compilation and maintenance of sequences of command line programs but also allows KNIME workflows to take advantage of the compute power of a LINUX cluster. RESULTS: Use of the command line programs is demonstrated in three different workflow examples: (1) A workflow to create a data file with project-relevant data for structure-activity or property analysis and other type of investigations, (2) The creation of a quantitative structure-property-relationship model using the command line programs via KNIME nodes, and (3) The analysis of strain energy in small molecule ligand conformations from the Protein Data Bank database. CONCLUSIONS: The chemalot and chemalot_knime packages provide lightweight and powerful tools for many tasks in cheminformatics. They are easily integrated with other open source and commercial command line tools and can be combined to build new and even more powerful tools. The chemalot_knime package facilitates the generation and maintenance of user-defined command line workflows, taking advantage of the graphical design capabilities in KNIME. Graphical abstract Example KNIME workflow with chemalot nodes and the corresponding command line pipe.

Atom-Atom-Path similarity and Sphere Exclusion clustering: tools for prioritizing fragment hits.

BACKGROUND: After performing a fragment based screen the resulting hits need to be prioritized for follow-up structure elucidation and chemistry. This paper describes a new similarity metric, Atom-Atom-Path (AAP) similarity that is used in conjunction with the Directed Sphere Exclusion (DISE) clustering method to effectively organize and prioritize the fragment hits. The AAP similarity rewards common substructures and recognizes minimal structure differences. The DISE method is order-dependent and can be used to enrich fragments with properties of interest in the first clusters. RESULTS: The merit of the software is demonstrated by its application to the MAP4K4 fragment screening hits using ligand efficiency (LE) as quality measure. The first clusters contain the hits with the highest LE. The clustering results can be easily visualized in a LE-over-clusters scatterplot with points colored by the members' similarity to the corresponding cluster seed. The scatterplot enables the extraction of preliminary SAR. CONCLUSIONS: The detailed structure differentiation of the AAP similarity metric is ideal for fragment-sized molecules. The order-dependent nature of the DISE clustering method results in clusters ordered by a property of interest to the teams. The combination of both allows for efficient prioritization of fragment hit for follow-ups. Graphical abstractAAP similarity computation and DISE clustering visualization.

A searching and reporting system for relational databases using a graph-based metadata representation.

Relational databases are the current standard for storing and retrieving data in the pharmaceutical and biotech industries. However, retrieving data from a relational database requires specialized knowledge of the database schema and of the SQL query language. At Anadys, we have developed an easy-to-use system for searching and reporting data in a relational database to support our drug discovery project teams. This system is fast and flexible and allows users to access all data without having to write SQL queries. This paper presents the hierarchical, graph-based metadata representation and SQL-construction methods that, together, are the basis of this system's capabilities.

DISE: directed sphere exclusion.

The Sphere Exclusion algorithm is a well-known algorithm used to select diverse subsets from chemical-compound libraries or collections. It can be applied with any given distance measure between two structures. It is popular because of the intuitive geometrical interpretation of the method and its good performance on large data sets. This paper describes Directed Sphere Exclusion (DISE), a modification of the Sphere Exclusion algorithm, which retains all positive properties of the Sphere Exclusion algorithm but generates a more even distribution of the selected compounds in the chemical space. In addition, the computational requirement is significantly reduced, thus it can be applied to very large data sets.

Process-driven information management system at a biotech company: concept and implementation.

While established pharmaceutical companies have chemical information systems in place to manage their compounds and the associated data, new startup companies need to implement these systems from scratch. Decisions made early in the design phase usually have long lasting effects on the expandability, maintenance effort, and costs associated with the information management system. Careful analysis of work and data flows, both inter- and intradepartmental, and identification of existing dependencies between activities are important. This knowledge is required to implement an information management system, which enables the research community to work efficiently by avoiding redundant registration and processing of data and by timely provision of the data whenever needed. This paper first presents the workflows existing at Anadys, then ARISE, the research information management system developed in-house at Anadys. ARISE was designed to support the preclinical drug discovery process and covers compound registration, analytical quality control, inventory management, high-throughput screening, lower throughput screening, and data reporting.