Podcast Summary
Using deep learning to discover new antibiotics: Researchers used machine learning to identify a promising new antibiotic candidate, Hallucin, from large databases, validating its potential through experiments.
The article by Jonathan Stokes, Regina Barsley, James Collins, and colleagues, published in Cell, presents a groundbreaking deep learning approach to identify new antibiotic drugs. The authors used two large databases to create a novel machine learning method and identified a promising candidate named Hallucin. This drug has excellent antibiotic properties and a distinct structure, which is crucial in the fight against antibiotic resistance. The study's significance lies in the experimental validation of the predictions, which is a landmark example in the field of predictive drug discovery. The challenge in identifying new antibiotics is not only scientific but also business-related. Antibiotics are complex to develop, and the most innovative ones are reserved for more serious cases, making the discovery and development of new antibiotics a significant challenge. The potential consequences of not finding new antibiotics are alarming, as we may soon run out of effective drugs to combat bacterial infections.
Machine learning revolutionizes antibiotic discovery: Machine learning models can predict antibiotic effects and democratize drug discovery, reducing costs and time, leading to a surge in novel antibiotics.
The business aspect plays a crucial role in the development of new drugs, especially antibiotics. The high cost and risk involved make it challenging for companies to invest without the promise of substantial returns. However, recent advancements in machine learning and new business models could lead to a shift towards discovering drugs for smaller indications, similar to how niche content creators have emerged in the entertainment industry. In their study, researchers used a machine learning model to predict the inhibitory effect of compounds on E. coli, training it on a small dataset and later applying it to a larger one. The model didn't rely on pre-existing chemical structure information but instead created new representations, called fingerprints, from scratch. This approach, which overcame the limitations of traditional methods, led to the discovery of a new antibiotic molecule. This research represents a significant advancement in the field, as it demonstrates the potential for machine learning to democratize drug discovery and make it more accessible to a broader scientific community. By reducing the cost and time required to bring new drugs to market, this technology could lead to a surge in the development of novel antibiotics, addressing the pressing issue of antibiotic resistance.
Discovering Halicin: A New Antibiotic through Deep Learning: Deep learning in drug design led to the discovery of halicin, an antibiotic with potent effects against various pathogens and unique structure, through a cost-effective process involving comprehensive experiments.
Deep learning approaches in drug design, like the discovery of the antibiotic halicin, allow the computer to infer and identify important features of a molecule, whereas in classical machine learning approaches, the researcher manually defines the features. Halicin, an antibiotic discovered using deep learning, is an attractive candidate due to its potent inhibitory effects on a wide range of pathogens, including E. coli and C. diff, and its ability to eradicate persistent cells. Additionally, it is structurally divergent from conventional antibiotics. The discovery of halicin involved comprehensive in vitro and in vivo experiments, including the disruption of the proton motor force, a distinct and new function being targeted. The full stack of experiments, from lab to animal models, is appealing due to the relatively small budget required compared to other areas of research like Alzheimer's. Deep learning in drug design helps us discover what we don't know, while classical machine learning focuses on what we already know. Halicin's discovery showcases the potential of deep learning in drug design and the importance of a complete, well-rounded story in scientific research.
Revolutionizing Drug Development with Machine Learning: Machine learning methods can revolutionize drug development by identifying targets, optimizing leads, and predicting multiple properties for a large number of molecules. AI's role in medicine is increasingly important.
Machine learning approaches, like the one discussed in the article, have the potential to revolutionize various aspects of drug development, beyond just initial discovery. This includes identifying targets, optimizing leads, and even predicting multiple properties for a large number of molecules. The beauty of these methods is their agnostic nature, meaning they can be applied to a wide range of systems. Furthermore, using a multitask learning framework can help regularize predictions and make them more robust. The article specifically identified new antibiotic candidates, including halicin, using deep neural networks for drug lead identification. This approach, which can shape the future of drug discovery and development, highlights the increasingly important role of AI in medicine. Another interesting topic discussed was the research on the novel coronavirus causing COVID-19. Two articles were mentioned: Walz et al.'s "Structure, Function, and Antigenicity of the SARS CoV-two Spike Glycoprotein" and Rapp et al.'s "Cryo EM Structure of the 2019 nCoV Spike in the Pre Fusion Confirmation." These studies provide crucial insights into the virus's structure and function, which can aid in the development of diagnostic tools, therapeutics, and vaccines. Understanding the spike glycoprotein's structure and function is essential for designing effective interventions against the virus. These studies are significant steps forward in the ongoing fight against COVID-19.
Comparing the Spike Proteins of SARS CoV-2 and SARS CoV: Both SARS CoV-2 and SARS CoV use ACE 2 for entry, but their spike proteins have differences in glycans and furin cleavage sites, impacting immune response and drug/vaccine development
Both SARS CoV-2 and SARS CoV, two coronaviruses responsible for different pandemics, have similarities and differences in their spike proteins. These proteins play a crucial role in the virus's interaction with the host cell, particularly in binding to the ACE 2 receptor and the presence of a furin cleavage site. Both viruses use ACE 2 for entry, but the spike proteins have subtle differences, including the number and type of glycans, which can impact how the immune system recognizes the virus. Understanding these similarities and differences is essential for developing drugs and vaccines, as targeting the spike protein is a primary strategy for combating these viruses. The presence of a furin cleavage site is a critical aspect of viral entry, and its presence or absence may contribute to the differing outcomes of these viruses. Overall, these studies highlight the importance of understanding the structural and functional differences between related viruses to effectively combat future outbreaks.
Unique feature of COVID-19 virus enhances its ability to infect cells: The COVID-19 virus, which causes SARS 2019, has a unique furin cleavage site that increases its ability to bind to the ACE 2 receptor and infect cells, contributing to its increased virulence. Ongoing research into how the virus interacts with the ACE 2 receptor could lead to potential treatments.
The SARS 2019 virus, which causes COVID-19, has a unique feature compared to the SARS 2,002 virus: a furin cleavage site. This site enhances the virus' ability to infect cells by increasing the probability of successful viral entry. The virus binds to the ACE 2 receptor, which is commonly found on human cell surfaces and is involved in regulating blood pressure. Both the 2019 SARS and 2002 SARS viruses use this receptor. However, the 2019 SARS virus binds to it with 10 to 20 times higher affinity, which may contribute to its increased virulence. Understanding how the virus interacts with the ACE 2 receptor could lead to potential treatments. The rapid publication of these studies highlights the importance of ongoing research in understanding the virus and developing effective countermeasures. Despite potential challenges, such as the critical role of ACE 2 in human physiology, continued investigation is crucial.
Two studies on SARS-CoV-2 show conflicting results on antibody cross-reactivity with SARS from 2002: While some studies suggest no cross-reactivity between antibodies against SARS from 2002 and SARS-CoV-2, others show cross-reactivity. The differences may be due to the use of polyclonal vs monoclonal antibodies and the specific regions of the spike protein they bind to.
While two research papers on the SARS-CoV-2 virus reached some concordant findings, such as identifying ACE 2 as the receptor and the similarity of the spike protein, they showed opposite results in the area of antibody cross-reactivity between SARS from 2002 and SARS-CoV-2. The Walls et al paper used polyclonal antibodies generated from mice, which bind to various regions of the spike protein and prevent viral entry into human cells. In contrast, Rapp et al used monoclonal antibodies, which are highly specific to a subdomain within the 2002 SARS spike protein and only measure binding. The findings of no binding in the Rapp et al study contrasted with the Walls et al study, which showed that polyclonal antibodies could prevent viral entry. However, it's important to note that these results are not necessarily mutually exclusive, as there could be a vast number of possible antibodies against these viruses, and the Rapp et al study only tested three monoclonal antibodies. Additionally, other studies support both the conclusion of no cross-reactivity and the conclusion of cross-reactivity. The ongoing research in this area highlights the importance of various approaches and techniques in understanding the SARS-CoV-2 virus.
New studies reveal protein structures using advanced techniques: Recent advancements in cryo EM technology allow scientists to determine protein structures quickly, providing valuable information for vaccine and therapeutic design. Both studies used in vitro methods, making them the fastest type of research in this field.
Both studies, despite using different techniques, managed to obtain similar results due to the recent advancements in quick data-generating scientific methods. Traditionally, determining the structure of high-quality proteins would require X-ray crystallography, a process that involves getting the protein to crystallize in a specific way, which is a challenging task. However, with the recent revolution in cryo EM, scientists can now determine protein structures by rapidly cooling the protein sample and shooting a beam of electrons at it. The resulting data can provide valuable information for vaccine and therapeutic design. Both studies were conducted in vitro, making them the fastest type of research in this field. As research continues, we can expect to gain new insights from experiments that take longer and involve more complex designs. The publication of these two studies side by side highlights the importance and value of the information that can be obtained in the shortest amount of time. Thanks, Judy, for discussing these articles with us. You can find all episodes of Journal Club at a16z.com. Thanks for listening.
