TechBio News: Drug Screening and AI

In vitro high-throughput screening and in silico methods for virtual screening: AI Tools and Startups

Drug Screening ?? and Chemical Space (Known and Virtual)

Searching for a new drug candidate, means searching and wandering around in a vast chemical space, comprising >10^60 molecules (…there are something like 10^22 to 10^24 stars in our Universe). In particular, the known chemical space—that includes public databases and corporate collections—probably contains something like 10^8 molecules (100 million), but the virtual chemical space might contain 10^60 compounds when considering only basic structural rules, or a more modest 10^20 – 10^24 molecules if combination of known fragments are considered.

Since the chemical space is far too large for an exhaustive "exploration", one is therefore left only with a partial, targeted "exploration" inside smaller virtual libraries and smaller chemical libraries.

Accordingly, numerous in silico methods are used to virtual screen compounds from smaller virtual chemical spaces along with in vitro high-throughput screening experiments of smaller chemical libraries, in order to identify drug candidates. As a matter of fact, the traditional drug discovery—known also as forward pharmacology—relies on the in vitro high-throughput screening (primary and secondary screening), while the rational drug design—also called reverse pharmacology or just drug design—relies on the computer modeling techniques and virtual screening (AI/ML tools and startups for rational drug design). From an experimental point of view, both traditional drug discovery and rational drug design are equally important.

Regarding the chemical libraries to screen, nowadays, several such chemical spaces are open access, including PubChem, ChemBank, DrugBank, ChemDB and more. When it comes to virtual libraries some solutions are:

• ?? A partnership with Enamine and Cresset will enable screening of ultra-large chemical spaces as part of the virtual screening process in drug discovery.
• ?? Discover eXplore: the largest virtual accessible and actionable Chemical Space providing over 7 trillion compounds for your novel drug discovery research by eMolecules.
• ?? PolarisQB announced integration of Virtual Libraries and Fragmentation schemes from chemical innovator Liverpool ChiroChem into Quantum-Aided Drug Design.
• ?? The Pan-Canadian Chemical Library: A Mechanism to Open Academic Chemistry to High-Throughput Virtual Screening.
• ?? Prepared Commercial Libraries | Schr?dinger.
• ? Virtual Screening - Drug Discovery - NUVISAN (NUVISAN library).
• ?? Virtual Compound Library Screening (by CD Computa Bio).
• ?? TargetMol: Compound Libraries | Inhibitors | Virtual Screening. And
• ?? CovalentInDB 2.0: an updated comprehensive database.

The logic behind the design of both types of libraries is often similar, and the two methods—experimental for real compound libraries and computational for virtual compound libraries—often complement each other in drug discovery. In the end, both types of libraries are commonly screened in parallel, and the results of one are compared to the other, aiming at discovering promising new drug leads.

AI During Drug Screening

Drug screening, known also as high-throughput screening or HTS, involves testing large chemical libraries of compounds on proteins, cells or animal embryos, and is divided in primary screening, that allows direct high throughput measurements on cells of chemical libraries, and secondary screening that is designed to confirm hits efficacy by a series of functional cellular assays. Currently, HTS robots can automate testing of 10^3–10^6 compounds of known structure per day.

For example, Revvity Inc (NYSE: RVTY) (PerkinElmer Health Science Inc. is Now Revvity) has a novel multimode plate reader, called the EnVision Nexus platform for demanding HTS applications, that enables researchers to screen millions of samples since it can be equipped with a plate stacker for 20 or 50 plates, and can be fully automated and integrated for 24/7 workflow-driven automation. They offer also Opera Phenix? Plus high-content imaging system that was designed for high-throughput high-content imaging assays, phenotypic screening, assays using complex disease models, such as live cells, primary cells and microtissues, and fast-response assays, such as Ca2+ flux. Moreover, mobile robots ?? by Biosero are already playing a critical role in supporting various scientific workflows, proving their value in high-throughput screening.

To give another example, Molecular Devices, one of the leading providers of high-performance bioanalytical measurement solutions for life science research, pharmaceutical and biotherapeutic development, introduced the CellXpress.ai? an automated cell culture system for screening, that is a revolutionary ML-assisted solution that standardizes the entire cell culture journey. The CellXpress.ai? is an AI-driven cell culture innovation hub that gives you total control over demanding cell culture feeding and passaging schedules—eliminating time in the lab while maintaining a 24/7 schedule for growing and scaling multiple stem cell lines, spheroids or organoids. Molecular Devices is also offering

• an AI-based software that provides Photoshop-like tools for image annotation, and
• the ImageXpress? Confocal HT.ai High-Content Imaging System, designed to help researchers advance phenotypic screening of 3D organoid models. The ImageXpress? utilizes a seven-channel laser light source with eight imaging channels to enable highly multiplexed assays while maintaining high throughput by using shortened exposure times. Water immersion objectives improve image resolution and minimize aberrations so scientists can see deeper into thick samples. Moreover, the combination of MetaXpress? software and IN Carta? software simplifies workflows for advanced phenotypic classification and 3D image analysis with ML capabilities and an intuitive user interface.

Just as I mentioned before, since it is impossible to investigate at once the vast chemical space most scientists prefer targeted screen inside smaller chemical libraries in public databases and corporate collections—containing something like 10^8 molecules—for example: PubChem, Chemspider, ZINC, NCI, ChemDB, BindingDB, ChEMBL, CTD, HMDB, SMPDB, DrugBank.

For instance, PubChem is the world's largest collection of freely accessible chemical information and ZINC20 is a free database of commercially-available compounds for virtual screening that contains over 230 million purchasable compounds in ready-to-dock, 3D formats. ZINC also contains over 750 million purchasable compounds you can search for analogs in under a minute.

Other databases of the known and unknown chemical space are: SuperScent (scents from literature), Flavornet (volatile compounds from literature), SuperSweet (carbohydrates and artificial sweeteners), BitterDB (bitter cpds from literature and Merck index), GDB-11 (molecules of up to 11 atoms of C, N, O, and F), GDB-13 (molecules of up to 13 atoms of C, N, O, S, and Cl) and GDB-17 (molecules of up to 17 atoms of C, N, O, S and halogens).

Screening and Imaging ??

During HTS screening, AI tools are mostly used for sorting and for image-based phenotyping after treatment (image-based profiling) and one can distinguish two approaches:

• screening applications (often called high-content) that are focused on a specific phenotype with the aim to identify drugs (or drug targets) that can modulate it (i.e. modulate the subcellular localisation of a specific protein). And
• global profiling of perturbations after cell treatment that is complementary to techniques like transcriptional profiling. For this reason, the sub-cellular structures are stained with fluorescent dyes or antibodies that are used to ‘paint’ and visualize cells and sub cellular structures, while automated image analysis is subsequently used to profile the phenotype of these cells.

In general, computer vision can extract multivariate feature vectors of cell morphology such as cell size, shape, texture and staining intensity without further human intervention.

For example, at SLAS 2023, the Tokyo-based ?? ThinkCyte announced the commercial launch of VisionSort, an AI enabled technology for characterizing and sorting cells after treatment based on image information at record high-throughput rates by integrating a novel ultrafast imaging technique with AI. In particular, VisionSort is the first cell sorter to combine analytical features found in conventional fluorescence flow sorters with the ability to perform label-free cell sorting and unbiased morphological analysis of cell populations. The combined capabilities can enable researchers to not only view cells based on a combination of known markers and morphological phenotypes, but also sort out unique populations for downstream processing or molecular analysis.

Moreover, apart the characterization of the morphological alterations of the cells and specific biomarkers associated to a novel drug candidate during the screening phase of drug discovery, also the toxicology profile ????? of the novel compound is also investigated in order to ensure good absorption, distribution, metabolism, excretion and toxicity (ADMET) properties (ADMET Prediction and AI: Startups for ADMET Prediction).

By way of illustration, when ADMET prediction met AI, the following tools were born:

• BBBper is a ML online tool for Blood-Brain Barrier (BBB) permeability prediction,
• ProTox 3.0 is a webserver for the prediction of toxicity of chemicals utilizing molecular similarity and ML models,
• VenomPred 2.0 is a ML based platform for molecular toxicity prediction,
• SolTranNet is a ML model for predicting aqueous solubility from a molecule’s SMILES representation,
• CYPlebrity is a ML model for predicting inhibitors of cytochrome P450 enzymes, and
• ToxTree is a ML based model to predict hERG and Nav1.5 cardiotoxicity of a molecular compound at the outset of drug development process.
• For more: ADMET predictions: different approaches, including ML models, for the prediction of specific ADMET endpoints.

A different problem during screening is assay interference ??? caused by small molecules. Several approaches have been developed that allow scientists to flag potentially “badly behaving compounds” or “bad actors” or “nuisance compounds”. Usually, these compounds are typically aggregators, reactive compounds and/or pan-assay interference compounds (PAINS), and many are frequent hitters.

The solution to this problem comes from Hit Dexter, a recently introduced ML approach that predicts how likely a small molecule is to trigger a false positive response in biochemical assays (including also the binding of compounds based on “privileged scaffolds” to multiple binding sites). The models used by Hit Dexter were derived from a dataset of 250,000 compounds with experimentally determined activity for at least 50 different protein groups.

The new Hit Dexter 2.0 web service covers both primary and secondary screening assays, providing user-friendly access to similarity-based methods for the prediction of aggregators and dark chemical matter (a set of drug-like compounds that has never shown bioactivity despite being extensively assayed), as well as a comprehensive collection of available rule sets for flagging frequent “bad” hitters and compounds including undesired substructures.

Several other prediction models have been constructed for public applications, including pan-assay interference compounds (PAINS), like: Aggregator Advisor, Luciferase Advisor, ALARM NMR, FAF-Drugs4, Badapple, Hit Dexter 3.0, Lilly-MedChem, ChemAGG, ChemFLuc and ChemFluo.

To give an example, ChemFH is an integrated online platform facilitating rapid virtual evaluation of potential false positives, including colloidal aggregators, spectroscopic interference compounds, firefly luciferase inhibitors, chemical reactive compounds, promiscuous compounds, and other assay interferences.

Let’s now see the startups involved in the HTS screening utilizing AI.

?? High-Throughput Screening Startups

Synsight (France ??)

• ? Synsight is a deep tech company developing a screening technology that enables the development of effective first-in-class drug candidates (for RNA targeting) based on an AI discovery platform and cell imaging, with phenotypic assays, high-content screening and high-content imager allowing to acquire more than 60000 images per day. In particular, Synsight developed the Microtubule Bench technology (MT bench?), an industrialised cell testing to screen molecules by microscopy and identify and quantify the modulations of small molecules on protein-protein interactions or between protein and nucleic acid.
• To target interactions between mRNA and RNA-binding proteins, MTBench? requires the fusion of a fluorescently tagged RNA-binding protein of interest with a microtubule-binding domain (MBD). When expressed in a cell, the fusion protein is directed onto the microtubules, where it can behave like a bait and bind prey mRNA. The MTBench? assay employs high-throughput microscopy to derive a correlation score from a single test condition, reflecting the colocalization between the fluorescence of baits and preys on the same microtubules. This score is then translated into a mathematical regression slope, serving as the assay output. In the MTBench? assay, a positive slope indicates an interaction, while a reduced slope value signifies the inhibition of this specific interaction in the presence of an active compound. This straightforward method allows for easy adaptation to target other interfaces, like interactions between proteins or between a protein and a specific RNA sequence.
• On July 08, 2024, Iktos (Paris, Ile-de-France, France, 2016) that is a company specializing in the development of AI solutions applied to chemical research (more specifically medicinal chemistry and new drug design) acquired Synsight.

Ailynix (Santa Clara, CA, US ??)

• ? Ailynix is an AI Drug Design and Discovery company developing tools in order to predict the biological activity of chemical compounds. For example, Zunomics, a subsidiary of Ailynix, has developed a Computational Antiviral Drug Discovery (CAViDD) Platform to discover novel oral antiviral drugs, by mining unexplored chemical space to deliver innovative medicines.

Pangea Bio (Tel Aviv, Israel ????)

• ? Pangea Bio utilizes AI to uncover promising molecules from nature’s diverse chemical space, enhanced by traditional human evidence of safety and efficacy. In particular, The PangeAI discovery platform (Knowledge Graph, Computational Metabolomics, Compound Activity Profiling) accelerates the discovery and development of novel therapeutics from plants and other kingdoms of life, to translate nature’s metabolome into medicine, for neurological and neuropsychiatric diseases.

PhenomicAI (Canada ??)

• ? PhenomicAI (2017) is using advanced ML tools for processing imaging, RNA sequencing and spatial transcriptomics data to understand the biology of single-cells sitting in complex multi-cell systems. This allows them to explore how cells communicate with each other, both in human tissue samples and disease relevant experimental models. On November 29, 2023, Phenomic and Boehringer Ingelheim announced a 500M collaboration to utilize Phenomic’s single-cell RNA computing platform, dubbed scTx, to unlock new targets.

Aiforia (Helsinki Finland ????)

• ? Aiforia (2013) by analyzing images uploaded to its cloud is allowing researchers to detect any visible feature or pattern at scale—including in tissues and cells—in order to understand pathophysiology. Aiforia’s platform brings together AI and high-performance cloud computing and assists image-based diagnostics by providing efficient and scalable solutions, enabling new preclinical discoveries and clinical support with highly accurate and consistent data.
• In 2022, Genuv (a clinical-stage biotechnology company focused on innovative drug discovery for the central nervous system disorders) announced an agreement with Aiforia, to add Aiforia’s AI for medical image analysis to its ATRIVIEW discovery platform. The ATRIVIEW AI platform is used to screen both existing drugs and new substances for neuroprotective and neurogenerative effects.
• Aiforia’s image analysis tools are also used by Sanofi, Boehringer Ingelheim, AstraZeneca and Bristol Myers Squibb, to help translate images into discoveries during screening.

VariationalAI (Canada ????)

• ? Variational AI in Vancouver (developer of the Enki? platform) leverages a powerful new form of ML known as generative AI to free scientists from reliance on screening and libraries (both experimental and virtual) and to eventually generate de novo molecules with all the optimized properties, in order to discover efficacious, safe and synthesizable small molecule therapeutics in a fraction of the time and cost. De novo is a more general term that refers to the greater category of methods that do not use templates as a prediction tool, instead they produce a series of possible candidate structures (called ‘decoys’) guided by scoring functions and sequence-dependent biases. Examples are: Creative Biolabs, De Novo Design Workflow By Schr?dinger, DENOPTIM, Creative Biostructure and Rosetta Commons.
• On September 17, 2024, Rakovina Therapeutics and Variational AI Announced a Drug Discovery Collaboration to jointly research DNA-damage response kinase targets to identify and develop novel small-molecule therapies against DNA-damage response (DDR) targets for the treatment of cancer. Rakovina (TSX.V: RKV) is developing Deep Docking, that is a novel DL platform that is suitable for docking billions of molecular structures in a rapid and accurate way by utilizing QSAR deep models trained on docking scores of subsets of a chemical library, to approximate the docking outcome for yet unprocessed entries and, therefore, to remove unfavorable molecules in an iterative manner. The use of the DD methodology in conjunction with the FRED docking program allowed rapid and accurate calculation of docking scores for 1.36 billion molecules from the ZINC15 library against 12 prominent target proteins and demonstrated up to 100-fold data reduction and 6000-fold enrichment of high scoring molecules, without notable loss of favorably docked entities (Deep Docking: A Deep Learning Platform for Augmentation of Structure Based Drug Discovery).
• During the Rakovina-Variational AI collaboration, Variational AI will employ the EnkiTM platform to identify novel inhibitors of specific DDR kinase targets selected by Rakovina Therapeutics, that will synthesize them and evaluate the viability of these drug candidates as potential cancer therapies in its laboratories at the University of British Columbia.

Terray Therapeutics (US ??)

• ? Terray Therapeutics (2018) has a small molecule screening (experimental and virtual) and optimization platform (tArray) for generative AI-driven drug discovery, integrating chemical experimentation and computation on an unprecedented scale. They explore molecules and targets more broadly and deeply with a sophisticated combination of ultra-high throughput experimentation, generative AI, biology, medicinal chemistry, automation and nanotechnology. Their experimental dataset includes more than two billion unique target-ligand binding measurements, growing by 150 million new measurements every month. They can screen 2M molecules against a target in four minutes and convert 25 TB of image data into binding affinities daily, allowing them to rapidly identify potent and selective molecules.
• On October 17, 2024, Terray Therapeutics announced it has raised $120M in a Series B fundraise to enhance its AI platform (Terray Therapeutics rakes in $120M for AI-powered small-molecule drug development). New investor Bedford Ridge Capital and existing investor NVentures, Nvidia Corp.’s venture capital arm, led the funding round. The round also attracted investments from Maverick Capital, Goldcrest Capital, Madrona Ventures, Two Sigma Ventures, XTX Ventures, Digitalis Ventures and Alexandria Ventures. This new financing brings the total raised by the company to more than $200M.
• Terray, Benchling, Dotmatics, TetraScience and Cadence Molecular Sciences (OpenEye), are all using NIM Agent Blueprints by NVDIA in their computer-aided drug discovery platforms. NVIDIA (NASDAQ: NVDA) released on August 27, 2024 the NIM Agent Blueprint for generative AI-based virtual screening (Better Molecules, Faster: NVIDIA NIM Agent Blueprint Redefines Hit Identification With Generative AI-Based Virtual Screening), that identifies and improves virtual hits in a smarter and more efficient way and has at its core three essential AI models:

AI-Driven Virtual Screening Approaches: Structure-based, Ligand-based and Chemogenomic

In general, the AI-Driven Virtual Screening for lead compound identification during drug design consists of:

1?? Structure-based approaches: molecular docking simulations (namely a computational technique that predicts the binding affinity of ligands to receptor proteins) that involves a two-step process of 1) conformational space search and 2) scoring. For example ??

• The Deep Docking (DD) platform enables up to 100-fold acceleration of structure-based virtual screening by docking only a subset of a chemical library, synchronized with a ligand-based prediction of the remaining docking scores. This AI method results in hundreds to thousands-fold virtual hit enrichment (without significant loss of potential drug candidates) and hence enables the screening of billion molecule–sized chemical libraries without using extraordinary computational resources.
• Regarding scoring during Structure-based Virtual Screening, traditional scoring functions, as well as data-driven machine learning (MLSF) and deep learning-based scoring functions (DLSF), such as 3D convolutional neural networks (3D-CNN) and graph convolutional networks (GCN), can prioritize ligand poses and estimate binding affinity.
• An essential step in the development of virtual screening methods is the use of established sets of actives and decoys for benchmarking and training, and DUD-E is designed to help benchmark molecular docking programs by providing challenging decoys. “Decoys” is essentially a codename for evaluating how well a docking program has done on a target and refers to a set of molecules that (probably) won't bind to a target. Some decoys datasets are: ?? DataSheet2_A Set of Experimentally Validated Decoys for the Human CC Chemokine Receptor 7 (CCR7) Obtained by Virtual Screening.xlsx. ?? FieldScreen: Virtual Screening Using Molecular Fields. Application to the DUD Data Set. ?? Data from: Deep Reinforcement Learning Enables Better Bias Control in Benchmark for Virtual Screening. ?? 21 Datasets.
• In general, the Structure-Based Drug Designing (SBDD) approach is the current mainstream mode for the rational drug design of small molecules and includes structure determination of the target protein, cavity identification, ligand database construction, ligand docking and lead discovery, combined with a computational-based virtual screening of large libraries. The softwares used for SBDD are: SWISS-MODEL, MODELER, Phyre and Phyre2, CASTp, Active site prediction tool, AutoDockVina and Schrodinger. A novel method just presented for SBDD is MISATO: a machine learning dataset of protein–ligand complexes for structure-based drug discovery. Another method just introduced, is SurfDock that is a surface-informed diffusion generative model for reliable and accurate protein–ligand complex prediction.
• However, LigPose is a novel geometric ????? DL technique that does not require docking and can effectively predict the native-like shape of ligands with matching protein targets and binding strengths. It entails modeling the ligand and its binding target as an entire undirected graph, where each atom is represented by each node. Atom coordinates in Euclidean space are used to optimize the three-dimensional structures; binding strength and correlation-enhanced graph learning are auxiliary tasks.

The AI-Driven Virtual Screening for lead compound identification during drug design, apart from the structure-based approaches, consists also of ligand-based approaches.

2?? Ligand-based approaches: they employ quantitative structure-activity relationship (QSAR) models, generating molecular descriptors to describe compounds, and ML models predict the bioactivity using these descriptors. These methods include: graph-based models (recurrent neural networks {RNNs}, neural graph fingerprints), sequential models (long short-term memory - a type of RNN for sequential compound representation) and similarity-based models (molecular fingerprints, transcriptomic expressions). Some examples are ??

• PyRMD is a A New Fully Automated AI-Powered Ligand-Based Virtual Screening Tool.
• On the relevance of query definition in the performance of 3D ligand-based virtual screening: an analysis was performed using PharmScreen, a 3D LBVS tool that relies on similarity measurements of the hydrophobic/philic pattern of molecules, and Phase Shape, which is based on the alignment of atom triplets followed by refinement of the volume overlap.
• Celeris Therapeutics is using ML to predict biomolecular interactions and generate new chemical entities. By using the Xanthos platform for design, they start with a protein target sequence, determine its 3D structures, predict ligand binding, generate linkers, predict ternary complexes and refine proximity-inducing compounds selection through several layers of filters including molecular dynamics and QSAR.
• CSAR DOCK is a benchmark database of protein-ligand complexes with various crystal structures and binding affinities.
• Moreover, QSAR based approaches have proven to be very valuable in predicting ADMET. QSAR modeling is a well-known technique that reveals associations between structural characteristics and biological or toxicological activities under the general assumption that similar chemical structures display similar activities (US FDA Experience in the Regulatory Application of (Q)SAR). In fact, QSAR studies are already used extensively and are also increasingly being accepted within the regulatory decision-making process as an alternative to animal tests for toxicity screening. In a recent effort, two QSAR models were developed to predict drug permeability across the blood brain barrier, to assist regulators with abuse liability assessment of drug metabolites or other materials related to drug substances (New developments in regulatory QSAR modeling: a new QSAR model for predicting blood brain barrier permeability).
• On top of that, the new MRA Toolbox v.1.0 is a web-based toolbox including four additive toxicity modules: two conventional (Concentration Addition and Independent Action) and two advanced (Generalized Concentration Addition and Quantitative Structure–Activity Relationship (QSAR)-based Two-Stage Prediction) models. In particular, these QSAR based models have proven to be very valuable in predicting physicochemical properties, biological activity, toxicity, chemical reactivity and metabolism of chemical compounds during screening.

Finally, apart the structure-based approaches and the ligand-based approaches, the AI-Driven Virtual Screening for lead compound identification during drug design, consists also of chemogenomic methods.

3?? Chemogenomic methods: combine target proteins and compounds to predict drug-target interactions (DTIs), focusing on similarities between proteins and compounds, or feature-based, using fixed-length feature vectors to describe targets and compounds. DL models, such as CNNs and deep belief networks, enhance feature-based methods.

• The chemogenomic screening has the potential to expedite the conversion of phenotypic screening projects into target-based drug discovery approaches. The immediate purpose of a chemogenomic screen is to characterize the effect that a set of small molecules has at the gene or protein level.
• The Structural Genomics Consortium (SGC) offers a variety of chemogenomic sets to support research in target discovery and biological mechanisms.
• Editorial: Computational chemogenomics: In silico tools in pharmacological research and drug discovery.

Additionally, the docking-based virtual screening, namely the identification of the location of the binding site of the drug candidate within the protein target, has traditionally be done by using the following types of methods to identify the potential druggable binding sites (link):

• Template based methods such as ? firestar, 3DLigandSite and Libra,
• Geometry based methods such as ? CurPocket, Surfnet and SiteMap,
• Energy based methods such as ? FTMap and Q-SiteFinder,
• As well as ML methods such as ? DeepSite freely available at www.playmolecule.org, Kalasanty, DeepCSeqSite and DeepVC being some of the newer approaches that are under rapid development in recent years. And In Silico Methods for Identification of Potential Active Sites of Therapeutic Targets: MetaPocket 2.0, COACH, PockDrug, FTMap, Sitemap.

Finally, in addition to the location of the binding site of the drug candidate within the protein target you have to evaluate its potential druggability that can be done for example using DoGSiteScorer, a web server that supports the prediction of potential pockets and gives a druggability estimation. DoGSiteScorer is an algorithm that detects pockets and estimates druggability by considering global and local pocket properties, by using support vector machines (supervised learning models with associated learning algorithms that analyze data for classification and regression analysis) to build a predictive model.

To conclude with druggability and algorithms, PocketVec is a novel approach to generate pocket descriptors via inverse virtual screening of lead-like molecules and performs comparably to leading methodologies while addressing key limitations. By systematically searching for druggable pockets in the human proteome, using experimentally determined structures and AlphaFold2 models, PocketVec identified over 32,000 binding sites across 20,000 protein domains. PocketVec descriptors were then generated for each site and an extensive similarity search was conducted, exploring over 1.2 billion pairwise comparisons. The results revealed druggable pocket similarities not detected by structure- or sequence-based methods, uncovering clusters of similar pockets in proteins lacking crystallized inhibitors and opening the door to strategies for prioritizing chemical probe development to explore the druggable space (Comprehensive detection and characterization of human druggable pockets through binding site descriptors).

Let’s move now to the giga-scale screening.

Giga-scale screening unleashes power of AI and virtual libraries

Thanks to AI and advanced computer resources, we now have the Giga large scale integration technology (VLS) for screening, that uses considerably larger initial chemical libraries (1010 to 1015) than the normal screening (Giga-scale virtual screening utilizing the V-SYNTHES MolSoft ICM-Pro approach). These advanced computational methods for VLS include fast flexible docking, modular fragment-based algorithms, DL models and hybrid approaches and are supported by rapid growth of affordable cloud computing, graphics processing unit (GPU) acceleration and specialized chips.

The key factors that have defined the recent advances in giga-scale screening are the structural revolution in biology, the rapid and marked expansion of the drug-like chemical space and the emerging computational approaches supported by the broad availability of cloud and GPU computing resources to support these methods at scale.

?? AI-Driven Virtual Screening Startups

Codexis (US ????)

• ? Codexis (NASDAQ: CDXS) is a leading enzyme engineering company that has a proprietary platform, the CodeEvolver, that provides in silico high-throughput assay screening. By using powerful ML tools, sophisticated molecular, cellular and bioanalytical workflows, at Codexis they can design and screen libraries of thousands of enzyme variants in high throughput, then sequence every variant and correlate its sequence with its performance in a highly application-relevant screen. Among Codexis' partners you can find
• Codexis, Inc.'s (NASDAQ:CDXS) high institutional ownership speaks for itself as stock continues to impress, up 17% over last week (December 14, 2024).

Daewoong Pharmaceuticals (South Korea ??)

• ? On February 21, 2024, the South Korean pharmaceutical giant Daewoong (Daewoong Pharmaceuticals Co., Ltd, ???? ????) (KRX: 069620) announced that it has completed its own new drug development system using AI, the Daewoong AI System (DAISY).
• In particular, they developed a collection of 800 million known compounds, called Daewoong Advanced Virtual Database, or DAVID, and a separate AI system called AI-based Virtual Screening (AIVS) tool to detect active substances targeting specific proteins based on 3D modeling technology. Utilizing both DAVID and AIVS, DAISY allows Daewoong researchers to discover new compounds and quickly predict drug properties, enabling also ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) research.
• HanAll Biopharma, Daewoong Pharmaceutical and NurrOn Pharmaceuticals Announce Successful Completion of a First-in-Human Study for Potential Disease-Modifying Therapy for Parkinson’s Disease (November 25, 2024).

Innophore (Austria ????)

• ? Innophore with a cutting-edge drug and enzyme discovery platform that uses AI guided point-cloud technology, not only analyses a protein’s 3D structure but includes extended surface properties (HALOS) and volumetric cavities (catalophores) to predict target's characteristics and reactivity in AI virtual screening simulations. Their AI-driven strategy to design novel therapeutic enzymes combines the Catalophore technology, a mix of prepared protein structural data (CATALObase), and search algorithms and patterns tailored to specific needs.
• Among their products you can find also Cavitomix a PyMol Plugin. CavitOmiX plugin for Schrodinger’s PyMOL, is a tool that allows you to analyze protein cavities from any input structure. You can dive deep into proteins, Catalophore cavities and binding sites using crystal structures and state-of-the-art AI models from OpenFold (powered by NVIDIA’s BioNeMo service), DeepMind`s AlphaFold and ESMFold by Meta. And by just entering any protein sequence users can get the structure predicted by OpenFold or ESMFold loaded into their PyMOL within seconds.

Delta 4 (Austria ??)

• ? Delta 4 conducts in silico screening prior to experimental screening. Delta4 leverages a proprietary computational analytics platform (e.valuation), combined with straightforward biomedical testing and clinical validation of candidate drugs. Their unique approach integrates big data in silico and experimental screens, offering the most efficient matching of clinical indications and compound/drug effect. And their focus is the repositioning of existing drugs for novel indications.

Pharmacelera (Spain ????)

• ? Pharmacelera (2015) is applying quantum theory to boost drug design via their two primary software packages: PharmScreen (for an accurate ligand-based virtual screening, 3D-LBVS), PharmQSAR (3D quantitative structure-activity relationship/QSAR tool that enables a combination of multiple fields of interaction in order to perform comparative molecular field analysis and comparative similarity indices analysis studies) and exaScreen (for novel synthesizable hits from a 3D limitless space).
• By utilizing MolXplore (New Graphical User Interface: MolXplore), scientists will be able to analyze the results of PharmScreen by loading the results and analyzing them according to different options, such as evaluating the similarity of the candidates with respect to the reference molecule, superimposing the molecular fields and the chemical structures, or filtering the compounds according to different physicochemical properties.
• On May 07, 2024, Pharmacelera and Enamine, the developer of the world’s largest and most reputable virtual space named REAL, announced the extension of their current partnership to explore an extraordinary magnitude of compounds, that has been extended by a 10 fold factor, when compared to the early version.
• Additionally, Pharmacelera developed a protocol integrating internal scripts and the PharmScreen? software for preparing 3D libraries tailored for LBVS and SBVS, that when applied to the Enamine Screening Collection (4.4M compounds), it delivers: 270M conformers for LBVS and 7.9M stereoisomers for SBVS. However, the protocol isn't just limited to Enamine’s collection since they have successfully extended it to other large commercial libraries, such as Molport Screening Compounds. For projects requiring a more targeted approach, their protocol can filter 3D libraries based on drug-like properties. With this standardized protocol and PharmScreen?’s cutting-edge capabilities, they offer one of the largest and most up-to-date 3D screening libraries.
• On the relevance of query definition in the performance of 3D ligand-based virtual screening (April 2014).

Deep Apple Therapeutics (US ??)

• ? Deep Apple Therapeutics is accelerating drug discovery with structure-based large library virtual screening and DL models. The San Francisco-based company founded in 2021 combines cryo-EM to explore receptor conformations (GPCRs and other target classes), DL and the docking of multi-billion compound libraries generated using the company's proprietary Orchard.ai? algorithm.
• Orchard.ai? is a virtual library expansion tool for ML enabled acceleration of hit discovery that generates project specific proprietary chemical libraries that are > 95% novel against Zinc22 (a free database of commercially-available compounds for virtual screening, that contains over 37 billion enumerated compounds in 2D), and significantly increases dock scoring against the target of interest, and increases drug likeness for hits with high synthetic tractability.
• Leveraging a powerful combination of cryo-EM enabled structural biology, AI-powered pocket extraction, molecular dynamics and protein movement modeling and ultra-high-volume internally generated virtual libraries, Deep Apple’s DL approach can quickly identify and advance novel candidates addressing well-validated biological targets for a broad range of diseases.
• Deep Apple was created by the life sciences venture capital Apple Tree Partners (ATP) with academic co-founders Georgios Skiniotis (a world leader in cryo-EM and GPCR structural biology at Stanford University), Brian Shoichet (a pioneer of virtual screening at University of California San Francisco) and John Irwin (the computational library authority who created the widely used ZINC free virtual library of more than 10 billion synthesizable compounds at University of California San Francisco).
• Deep Apple emerged from stealth in 2023 with a Series A commitment of $52M from ATP. Spiros Liras, Ph.D., is the founding CEO of Deep Apple and a Venture Partner at ATP.

Remedium (Canada ??)

• ? Remedium uses AI to reduce dependence on mass screenings?— both virtual and chemical?—?through rapid selection of small molecule agonists, antagonists, or functional mimics of protein drug candidates. The company is using ML techniques, capable of analysing any protein with known or suspected therapeutic value with great accuracy.

Qubigen (Presagen has merged with Qubist to become Qubigen)

• Qubigen is an exciting new venture (November 2024) that merges Presagen’s Federated AI technology and Qubist’s virtual screening and AI drug design capability. Qubigen’s mission is to create the largest global federated proprietary dataset and computational toolbox for AI driven drug design.

Gandeeva (Canada ????)

• ? Gandeeva’s technology includes three proprietary platform modules working in concert: 1) SPOTLIGH, a Target Selection Engine, an AI-based approach to identify a continuous stream of validated targets, 2) HYPERFOCUSTM, a Cryo-EM Engine, a state-of-the-art atomic resolution imaging to map druggable sites and 3) CRYO-CADDTM, a Drug Discovery Engine, a rapid iterative cycle to generate structural insights at the speed of chemistry.
• On March 30, 2023, Gandeeva Therapeutics (a UBC spin-off company led by Faculty of Medicine professor Dr. Sriram Subramaniam) announced the initiation of a research collaboration with Moderna Inc. to explore applications of Gandeeva’s technology platform for a Moderna program.

DeepCure (US ????)

• DeepCure (2018) is a startup that uses DL to discover small molecule therapeutics. Their proprietary PocketExpander? uses AI and physics-based methods to identify physical and chemical features on the protein surface that can interact with small molecules. The technology includes quantum mechanical simulations to calculate binding energies and molecular dynamics simulations of protein folding.
• MolDBTM, their proprietary database for virtual screening has over a trillion unique molecules. After identifying the most promising small molecule per target, they optimize for ADMET properties. Their patent-pending molecular generation tool, MolGen?, leverages state of the art deep reinforcement learning to construct synthesizable molecules that efficiently capture all causal interactions for binding and selectivity while maintaining the desired ADME-tox profile (DeepPropR?).
• MolGen? generates custom libraries of compounds that are designed to selectively interact with features identified by PocketExpander?. The process leverages state-of-the-art deep reinforcement learning (RL) to build readily synthesizable molecules using millions of available building blocks and more than 200 robust chemical reactions. MolGen? also ensures that each molecule has the desired ADME-Tox profile and target candidate profile (TCP). MolGen? generates novel compounds that are readily synthesizable and can selectively interact with the features identified by PocketExpander? as well as meet ADMET and TCP requirements.
• On January, 4, 2023, Biosero, a developer of laboratory automation solutions, and DeepCure announced their collaboration on an advanced, fully automated chemistry synthesis platform to accelerate drug discovery. On May 28 2024, DeepCure announced that its Inspired Chemistry? platform had achieved a breakthrough in chemical synthesis synthesizing nirmatrelvir and 56 analogs in parallel using a robot-driven workflow.

Until next time ??????