This article is original from Translational Medicine Network. Please indicate the source when reprinting
Author: Lily
Introduction : Recently, Professor Yang Shengyong, the National Key Laboratory of Biological Treatment at West China Hospital of Sichuan University, , Professor Yang Shengyong, , published the latest research results. This study reports the research process of discovering RIPK1 small molecule inhibitors based on deep learning; it shows the ability of generative deep learning (GDL) models to generate completely new molecular structures, indicating that deep learning has great potential in the field of drug discovery.
Discovering Challenges of Brand New Skeleton Structure Active Compounds
01
Developing new drugs is an expensive and time-consuming process that can take more than $1 billion and 10 years. In the early stages of drug development, especially for innovative drug development, it is a crucial step for to efficiently discover emergent compounds or lead compounds with novel backbone structures. The traditional strategy is to use high-throughput screening methods to screen existing compound libraries; however, because of the limited structural diversity of the existing compound library and the repeated screening of the compound library by major pharmaceutical companies and drug R&D institutions, it may not be possible to find other active compounds with different scaffolds, and discovering new framework structural active compounds with independent intellectual property rights has become increasingly challenging.
At present, people have proposed to solve the above problem through de novo molecular design - the generation of new molecules with the required properties by computing. However, traditional methods of de novo molecular design, including structure-based molecular design methods, ligand-based and pharmacophore-based model-based methods involve a relatively manual process that requires experienced designers and clear design rules. Furthermore, since the head molecule design method is mainly fragment-based, the quality and diversity of the molecules it generates depends to a large extent on the fragment library and the algorithms used for fragment assembly.
In recent years, research on generative models or generative deep learning (GDL) models based on deep learning has achieved rapid development. Among them, the research on GDL model based on the recurrent neural network (RNN) is the most widely reported; while the conditional recurrent neural network (cRNN) can explicitly guide the subsequent molecular generation process by giving the RNN initial state vector as a condition.
However, existing cRNN and other GDL models still have many shortcomings - such as over-dependence on objective functions, generation of molecular novelty, limited diversity, etc. Furthermore, although most GDL models have been proven at the theoretical level, few examples of application to actual innovative drug discovery and success are still available.
New cRNN molecular generation model
Discover selective RIPK1 small molecule inhibitor
02
To solve the above problems, Professor Yang Shengyong 's team at the Biological Therapy Research Center of West China Hospital proposed a new cRNN molecular generation model - this model integrates strategies such as transfer learning, regularization enhancement and sampling enhancement. The research team successfully discovered RIPK1 kinase inhibitors using this strategy.
https://www.nature.com/articles/s41467-022-34692-w
RIPK1 is an serine /threonine protein kinase, which is involved in various signaling pathways for cell survival. It is worth noting that RIPK1 is also a key regulator of programmed cell necrosis (necroptosis), and is therefore closely related to the occurrence and development of various inflammatory and immune diseases.
When necrotic apoptosis is triggered by stimuli such as the tumor necrosis factor family, RIPK1 will be activated first. The activated RIPK1 binds to its downstream protein RIPK3, which then recruits and phosphorylates pseudokinase mixed lineage kinase domain (MLKL) phosphorylated MLKL to form oligomers and transfers to cell membrane to perform necrotic apoptosis. is based on the above-mentioned core role of RIPK1 in necrotizing apoptosis, and it is considered a promising target for the treatment of necrotizing apoptosis-related diseases.
By establishing a customized library of virtual compounds, virtual screening, chemical synthesis and biological activity verification of RIPK1 inhibitors, the researchers obtained a highly active and selective RIPK1 inhibitor (RI-962) without any modification. Subsequently, the researchers analyzed the crystal structure of RIPK1–RI-962, and structurally elucidated the molecular mechanism of RI-962's high activity and selectivity. In addition, the team also evaluated the in vivo effect of RI-962 on a mouse model of TNFα induced systemic inflammation response syndrome (SIRS) and DSS-induced inflammatory bowel disease (IBD); results showed that RI-962 improved TNFα-induced SIRS and DSS-induced IBD damage by inhibiting RIPK1 kinase activity. The position of the selected molecule of
for further experimental verification in TMAP of the filtered molecule:
Center: Overview of TMAP colored by the number of RECAP fragments.
surrounding box: a larger version of its corresponding field, where the molecules are colored by docking fractions (red-yellow-green).
molecules with docking fractions: selected molecules point to their position in TMAP.
Research significance
03
The novel conditional recurrent neural network (cRNN) molecular generation model developed by Professor Yang Shengyong's team has been successfully applied to establish a virtual compound library for RIPK1. The generated libraries are richer in than known RIPK1 inhibitors. Through a standard drug screening process against established compounds libraries, the research team retrieved a potent selective RIPK1 inhibitor, which inhibitor has previously not reported stents.
On the one hand, this application example verifies the effectiveness of this generative deep learning (GDL) model. Although RIPK1 is an kinase , this GDL model can be applied to different types of biological targets—the only requirement is that the biological target must have a sufficient number of known active compounds (target data). The more active compounds are known, the better the performance of the GDL model will be.
On the other hand, this application example leads to the identification of a potent RIPK1 inhibitor with previously unreported stents (RI-962). It is worth noting that RI-962 shows high selectivity for other 407 kinases. It also shows effective activity in vitro and in vivo. Even so, the compound still has some adverse properties that need further optimization in the future.
Reference materials:
https://www.nature.com/articles/s41467-022-34692-w
http://www.wchscu.cn/academic/70225.html
Note: This article aims to introduce the progress of medical research and cannot be used as a reference for treatment plans. If you need health guidance, please go to a regular hospital for treatment.
sub-forum 1
tumor immunity "Pujiang On the Dao" Schedule
sub-forum 2
tumor immunity "Songjiang Traceability" Schedule
Recommended·Live broadcast/Activity
11 November 18 14:00-17:30 Shanghai
Cell and Gene Therapy R&D and Industrialization Salon
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-