Research improves molecular probe of protein binding sites for drug discovery

Good news!

"... The invention works with an existing lab method called photo-crosslinking. Leaving behind a clean, uniform chemical signature, the technology allowed the team to directly compare how different molecules compete for the same binding site on a protein, all in a single experiment. Because most small-molecule drugs act by binding to specific protein targets, finding precisely where these molecules bind is a major benefit for drug discovery.

As proof of concept, the team analyzed the activity of dasatinib and ascinimib, two cancer drugs that target different sites on the same protein, a type of enzyme called a kinase that, when mutated, causes leukemia.

The results coincided with known interactions for each drug and revealed previously unknown interactions. The newer drug, ascinimib, which has a more favorable safety profile and fewer side effects, showed fewer off-target kinase interactions. ...

new technology, called SEE-CITE, is giving the molecule being studied the ability to detach from its payload so that each tagged molecule leaves behind a consistent calling card. This makes possible quantitative measurements and comparisons of how strongly different molecules engage a given binding site. The team also upgraded a widely used software tool to better interpret the complex data this method generates. ..." 

From the abstract:

"For chemical probe and drug discovery campaigns, the pairing of mass spectrometry-based chemoproteomics with photoaffinity labelling has emerged as a favoured approach for target discovery and mode of action assignment. However, photocrosslinked peptide-compound adducts raise analytic challenges for quantitative binding site discovery.

Here, to address these challenges, we establish the Silyl Ether Enables Chemoproteomic Interaction and Target Engagement (SEE-CITE) method. SEE-CITE incorporates a fully functionalized chemically cleavable photocrosslinking handle that enables precise site-of-labelling identification and head-to-head comparisons of relative binding site engagement by chemically diverse compounds.

To ensure high-confidence localization of labelled residues, we extended the MSFragger algorithm of the FragPipe computational platform to report localization scores customized for photoaffinity labelling and SEE-CITE data.

When applied to scout fragments and analogues of select FDA-approved kinase inhibitors, SEE-CITE delineates known drug binding sites and uncovers small-molecule binding sites that affect the protein activity of RTN4 and COX5A."

UCLA research improves molecular probe for drug discovery | UCLA

Small-molecule binding-site discovery using silyl ether-enabled chemoproteomics (open access)

Fig. 1: Establishing the SEE-CITE interaction site mapping platform using scout SEE-CITE probes.

Fig. 3: SEE-CITE mapping of ABL1 binding sites.

Africa launches a continental drug regulator, the African Medicines Agency (AMA)

Good news!

"Africa has inaugurated its own drug regulator, the African Medicines Agency (AMA), after ten years in the works. The agency aims to make drug development more relevant to African populations, whose genetic diversity is rarely reflected in global research, and harmonize regulation across its nations. The continent also faces rapid population growth and urbanization, and an upheaval in international funding. ..."

"After more than a decade of planning, the launch of the African Medicines Agency (AMA) is being celebrated in Mombasa, Kenya, this week at the Seventh Biennial Scientific Conference on Medical Products Regulation in Africa. The agency’s establishment marks a pivotal moment in Africa’s public health, at a time when the need for biomedical research conducted in Africa, focused on African health problems, has never been greater. ..."

Nature Briefing: Translational Research

Africa finally has its own drug-regulation agency — and it could transform the continent’s health "If it gets things right, the first major regulator of medicines to launch for 30 years could empower Africa to tackle African challenges around health and disease."

Failure is not an option for Africa’s newly launched medicines agency "The inequitable distribution of vaccines during the COVID-19 pandemic was the final proof of the need for more home-grown manufacturing and regulatory capacity across Africa."

African Medicines Agency (AMA)

Google's Isomorphic Labs announces $600m external investment round

Good news! Let AI dazzle us with new drugs!

"... Isomorphic Labs, a division within Google using AI to accelerate drug design, has raised $600 million from outside investors. Jared Kushner-owned Thrive Capital led the round ..."

Isomorphic Labs announces $600m external investment round - Isomorphic Labs

Credits: Techcrunch newsletter

RNA-editing drugs enter clinical trials to treat various genetic diseases

Good news!

"Drugs that edit mRNA sequences are beginning to enter clinical trials to treat various genetic diseases. Many of these use a family of enzymes called ADARs, which can repair or disable defective genes by swapping individual bases in their mRNA transcripts. Like its high-profile relative CRISPR, mRNA editing uses engineered nucleotide strings to lure the enzymes to their target genetic sequences. At least ten companies have launched or are about to launch early stage trials to treat the hereditary disorder alpha-1 antitrypsin deficiency, which is caused by a single mutation that is easily reversible through mRNA editing. In preclinical studies, one company’s editing system fixed up to 75% or more of the defective proteins."

"Oligonucleotide-based drugs already come in many flavours. The newest of these now aims to edit mRNA one base at a time, by harnessing endogenous enzymes called adenosine deaminases acting on RNA (ADAR). ...

A key difference between DNA- and RNA-editing is the dosing schedule. Whereas CRISPR-based drugs aim to provide a one-time treatment, the ADAR pipeline is mostly comprised of redosable drugs that are given every few weeks or less frequently. Their effects, as a result, are reversible. ..."

Nature Briefing

RNA-editing drugs advance into clinical trials ADAR-based editors that can change the mRNA code offer new opportunities in both rare genetic diseases and common complex ones.

New machine learning techniques boost predictions for virtual drug screening with less data using kernel methods

Recommendable! Sounds promising!

"... They came up with the first “transfer learning” techniques for kernel methods that can be successfully applied to large-scale datasets. ...
The researchers analyzed performance of their transfer learning algorithms on two massive Broad datasets, one from the Connectivity Map (CMAP) and the other from the Cancer Dependency Map (DepMap). These datasets describe the effects of drugs on cancer cell lines  across millions of drug and cell line combinations. The team trained their kernel method algorithms to predict either the genes expressed by a certain cell type after it was treated with a certain drug (using the CMAP dataset), or the proportion of cancer cells that survived after treatment with the same drug (using the DepMap dataset). ... Two additional advantages  of kernel methods are that they provide interpretability as well as a quantification of how uncertain the model is on a given prediction. ..."

From the abstract:

"Transfer learning refers to the process of adapting a model trained on a source task to a target task. While kernel methods are conceptually and computationally simple models that are competitive on a variety of tasks, it has been unclear how to develop scalable kernel-based transfer learning methods across general source and target tasks with possibly differing label dimensions. In this work, we propose a transfer learning framework for kernel methods by projecting and translating the source model to the target task. We demonstrate the effectiveness of our framework in applications to image classification and virtual drug screening. For both applications, we identify simple scaling laws that characterize the performance of transfer-learned kernels as a function of the number of target examples. We explain this phenomenon in a simplified linear setting, where we are able to derive the exact scaling laws."

New machine learning techniques boost predictions for virtual drug screening with less data | Broad Institute ... researchers have found a way to improve kernel methods, which are a simpler alternative to neural networks, for biomedical applications. 

Transfer Learning with Kernel Methods (open access)

Fig. 1: Our framework for transfer learning with kernel methods for supervised learning tasks.

AI-driven techniques reveal new targets for drug discovery associated with protein phase separation

Good news!

"The research team ... presented an approach to identify therapeutic targets for human diseases associated with a phenomenon known as protein phase separation, a recently discovered phenomenon widely present in cells that drives a variety of important biological functions.

Protein phase separation at the wrong place or time could disrupt key cellular functions or create aggregates of molecules linked to neurodegenerative diseases. It is believed that poorly formed cellular condensates could contribute to cancers and might help explain the aging process.

researchers ... developed a method for finding new targets for drug discovery in diseases caused by dysregulation of the protein phase separation process. The team found that they could replicate disease characteristics in cells by controlling the behaviour of these targets. ..."

From the significance and abstract:

"Significance

Given the emerging association between human disease and the protein phase separation (PPS) process, it is of great interest to identify possible targets for therapeutic interventions based on PPS regulation. This goal, however, is challenging because our understanding of the molecular origins of PPS-associated diseases is still in its infancy. Here, we present an approach based on the intersection of two components: 1) an analysis of a variety of parameters commonly used to identify therapeutic targets through the multiomic PandaOmics platform and 2) an assessment of the propensity of proteins to undergo PPS through the FuzDrop method. We apply this approach to prioritize human diseases for PPS-based interventions, and illustrate its implementation in the case of Alzheimer’s disease.

Abstract

The phenomenon of protein phase separation (PPS) underlies a wide range of cellular functions. Correspondingly, the dysregulation of the PPS process has been associated with numerous human diseases. To enable therapeutic interventions based on the regulation of this association, possible targets should be identified. For this purpose, we present an approach that combines the multiomic PandaOmics platform with the FuzDrop method to identify PPS-prone disease-associated proteins. Using this approach, we prioritize candidates with high PandaOmics and FuzDrop scores using a profiling method that accounts for a wide range of parameters relevant for disease mechanism and pharmacological intervention. We validate the differential phase separation behaviors of three predicted Alzheimer’s disease targets (MARCKS, CAMKK2, and p62) in two cell models of this disease. Overall, the approach that we present generates a list of possible therapeutic targets for human diseases associated with the dysregulation of the PPS process."

AI-driven techniques reveal new targets for drug discovery | University of Cambridge Researchers have developed a method to identify new targets for human disease, including neurodegenerative conditions such as Alzheimer’s disease.

Multiomic prediction of therapeutic targets for human diseases associated with protein phase separation (open access)

Fig. 2 Disease landscape representing opportunities for PPS-based therapeutics for 64 diseases across 10 disease groups.

IBM: An AI foundation model that learns the grammar of molecules

Very impressive! Google is not the only company fast advancing machine learning  for chemistry! The prospects of this research are great! This is only the beginning!

"... Introducing MoLFormer-XL, the latest addition to the MoLFormer family of foundation models for molecular discovery. MoLFormer-XL has been pretrained on 1.1 billion molecules represented as machine-readable strings of text. From these simple and accessible chemical representations, it turns out that a transformer can extract enough information to infer a molecule’s form and function. ...
We found that MoLFormer-XL could predict a molecule’s physical properties, like its solubility, its biophysical properties, like its anti-viral activity, and its physiological properties, like its ability to cross the blood-brain barrier. It could even predict quantum properties, like a molecule’s bandgap energies, an indicator of how well it converts sunlight to energy. ...
Many molecular models today rely on graph neural network architectures that predict molecular behavior from a molecule’s 2D or 3D structure. But graph models often require extensive simulations or experiments, or use complex mechanisms, to capture atomic interactions within a molecule. Most graph models, as a result, are limited to datasets of about 100,000 molecules, sharply limiting their ability to make broad predictions. ...
we trained MoLFormer-XL to focus on the interactions between atoms represented in each SMILES string through a new and improved type of rotary embedding. Instead of having the model encode the absolute position of each character in each string, we had it encode the character’s relative position. This additional molecular context seems to have primed the model to learn structural details that make learning downstream tasks much easier. ...
To pack more computation into each GPU, we chose an efficient linear time attention mechanism and sorted our SMILES strings by length before feeding them to the model. Together, both techniques raised our per-GPU processing costs from 50 molecules to 1,600 molecules, allowing us to get away with 16 GPUs instead of 1,000. By eliminating hundreds of unnecessary GPUs, we consumed 61 times less energy and still had a trained model in five days. ..."

From the abstract:

"Models based on machine learning can enable accurate and fast molecular property predictions, which is of interest in drug discovery and material design. Various supervised machine learning models have demonstrated promising performance, but the vast chemical space and the limited availability of property labels make supervised learning challenging. Recently, unsupervised transformer-based language models pretrained on a large unlabelled corpus have produced state-of-the-art results in many downstream natural language processing tasks. Inspired by this development, we present molecular embeddings obtained by training an efficient transformer encoder model, MOLFORMER, which uses rotary positional embeddings. This model employs a linear attention mechanism, coupled with highly distributed training, on SMILES sequences of 1.1 billion unlabelled molecules from the PubChem and ZINC datasets. We show that the learned molecular representation outperforms existing baselines, including supervised and self-supervised graph neural networks and language models, on several downstream tasks from ten benchmark datasets. They perform competitively on two others. Further analyses, specifically through the lens of attention, demonstrate that MOLFORMER trained on chemical SMILES indeed learns the spatial relationships between atoms within a molecule. These results provide encouraging evidence that large-scale molecular language models can capture sufficient chemical and structural information to predict various distinct molecular properties, including quantum-chemical properties."

An AI foundation model that learns the grammar of molecules | IBM Research Blog Meet MoLFormer-XL, a pretrained AI model that infers the structure of molecules from simple representations, making it faster and easier to screen molecules for new applications or create them from scratch.

Large-scale chemical language representations capture molecular structure and properties (no public access)

BenchSci Launches AI Tool to Map Disease Biology for Preclinical Drug Discovery

This seems to be a promising approach for faster drug development at lower cost!

"... BenchSci has launched a new AI software that aims to expedite preclinical phase drug development pipelines by extracting biological insights underlying disease.

The end-to-end SaaS (software as a service) platform, ASCEND, enables the discovery of biological connections, reduces unnecessary experiments, and uncovers risks at early stages. ... uses .. machine-learning technology to extract experimental evidence from secure internal and open external sources. The platform uses curated ontology datasets and compares experimental outcomes. This enables the software to create an evidence-based “map” of the biological mechanisms underlying different diseases.

... guide preclinical research by enhancing target selection, conducting due diligence, generating hypotheses, developing optimal investigative approaches, designing experiments, and identifying safety and efficacy risks to support IND (investigational new drug) submissions for clinical translations.

... incorporates publicly available scientific data from over 15 million published experiments and proprietary data generated by client companies that is securely available to respective customers. This approach helps R&D scientists understand the biological feasibility of new or existing lines of investigation and identify optimal approaches for testing hypotheses. ..."

BenchSci Launches AI Tool to Map Disease Biology for Preclinical Drug Discovery

Artificial intelligence model finds potential drug molecules a thousand times faster

Good news! And this is only the beginning! This is possibly a breakthrough! This seems to be an impressive work (Note: I have not studied this paper).

"... In a paper that will be presented at the International Conference on Machine Learning (ICML), MIT researchers developed a geometric deep-learning model called EquiBind that is 1,200 times faster than one of the fastest existing computational molecular docking models, QuickVina2-W, in successfully binding drug-like molecules to proteins. EquiBind is based on its predecessor, EquiDock ...
What’s more, 90 percent of all [newly developed] drugs fail once they are tested in humans due to having no effects or too many side effects. ..."

From the abstract:

"Predicting how a drug-like molecule binds to a specific protein target is a core problem in drug discovery. An extremely fast computational binding method would enable key applications such as fast virtual screening or drug engineering. Existing methods are computationally expensive as they rely on heavy candidate sampling coupled with scoring, ranking, and fine-tuning steps. We challenge this paradigm with EquiBind, an SE(3)-equivariant geometric deep learning model performing direct-shot prediction of both i) the receptor binding location (blind docking) and ii) the ligand's bound pose and orientation. EquiBind achieves significant speed-ups and better quality compared to traditional and recent baselines. ... Finally, we propose a novel and fast fine-tuning model that adjusts torsion angles of a ligand's rotatable bonds based on closed-form global minima of the von Mises angular distance to a given input atomic point cloud, avoiding previous expensive differential evolution strategies for energy minimization."

Artificial intelligence model finds potential drug molecules a thousand times faster | MIT News | Massachusetts Institute of Technology A geometric deep-learning model is faster and more accurate than state-of-the-art computational models, reducing the chances and costs of drug trial failures.

EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction (open access. The senior author, i.e. Tommi Jaakkola, is a well known ML researcher with over 40,000 lifetime citations)

Google creates Isomorphic Labs

Good news! Google's alphabet soup gets richer! Expect great things to happen!

None other than Demis Hassabis is the founder of the latest addition to Alphabet!

"I believe we are on the cusp of an incredible new era of biological and medical research. Last year DeepMind’s breakthrough AI system AlphaFold2 was recognised as a solution to the 50-year-old grand challenge of protein folding, capable of predicting the 3D structure of a protein directly from its amino acid sequence to atomic-level accuracy. This has been a watershed moment for computational and AI methods for biology.
Building on this advance, today, I'm thrilled to announce the creation of a new Alphabet company –  Isomorphic Labs – a commercial venture with the mission to reimagine the entire drug discovery process from first principles with an AI-first approach and, ultimately, to model and understand some of the fundamental mechanisms of life. ..."

Isomorphic Labs | Blog We are reimagining the entire drug discovery process from first principles with an AI-first approach

Credits: Andrew Ng's The Batch newsletter

Faster drug discovery through machine learning

Good news! Recommendable!

"... The new technique, dubbed DeepBAR, quickly calculates the binding affinities between drug candidates and their targets. The approach yields precise calculations in a fraction of the time compared to previous state-of-the-art methods. The researchers say DeepBAR could one day quicken the pace of drug discovery and protein engineering. ...

The affinity between a drug molecule and a target protein is measured by a quantity called the binding free energy ...

Methods for computing binding free energy fall into two broad categories, each with its own drawbacks. One category calculates the quantity exactly, eating up significant time and computer resources. The second category is less computationally expensive, but it yields only an approximation of the binding free energy. Zhang and Ding devised an approach to get the best of both worlds. ...
The “BAR” in DeepBAR stands for “Bennett acceptance ratio,” a decades-old algorithm used in exact calculations of binding free energy. Using the Bennet acceptance ratio typically requires a knowledge of two “endpoint” states (e.g., a drug molecule bound to a protein and a drug molecule completely dissociated from a protein), plus knowledge of many intermediate states (e.g., varying levels of partial binding) ...

DeepBAR slashes those in-between states by deploying the Bennett acceptance ratio in machine-learning frameworks called deep generative models. “These models create a reference state for each endpoint, the bound state and the unbound state,” ... These two reference states are similar enough that the Bennett acceptance ratio can be used directly, without all the costly intermediate steps. ...
“We’re sort of treating each molecular structure as an image, which the model can learn. ... “But here we have proteins and molecules — it’s really a 3D structure. So, adapting those methods in our case was the biggest technical challenge we had to overcome.” ..."

Faster drug discovery through machine learning | MIT News | Massachusetts Institute of Technology New technique speeds up calculations of drug molecules’ binding affinity to proteins.

Here is the link to the underlying research paper:

DeepBAR: A Fast and Exact Method for Binding Free Energy Computation (behind paywall)