Intelligence is an expression used to indicate an individual’s ability to comprehend complex situations and solve problems. It has always been the primary distinction between human beings and other life forms on earth. Intelligence is what enables humans to be aware and react to their environment. Over time, as mankind evolved, so did human intelligence and our desire to acquire knowledge. Today, the greatest minds of the world are not only unlocking the secrets of the universe and discovering cures for lethal diseases, but also seeking ways to further enhance their intelligence. To summarize, intelligence has three components, namely analytical, creative and practical. According to Sternberg, for an entity to be considered intelligent, it must be able to balance all the aforementioned attributes and be capable of using them to solve problems.
The intelligence of humans has enabled them to surpass physical barriers and tap into the reservoir of mental ability to create machines to simplify tasks and solve problems that were impossible for them to solve. Throughout history, humans have created innovative solutions to tackle various problems. For instance, humans overcame the problem of carrying heavier weights by inventing the pulley. We revolutionized the transport industry with the invention of the steam engine, which resulted in several applications across different industries. Similarly, problems associated with storage of data, data management and data analytics are now being actively tackled with several innovations in the computing domain. All these examples define the human ability to deal with complex problems through analysis, creativity and innovation. The global deep learning in drug discovery market is anticipated to grow at a CAGR of around 22.7%, till 2035, according to Roots Analysis. Driven by the ongoing pace of innovation and the profound impact of implementation of such solutions, deep learning is anticipated to witness substantial growth in the foreseen future.
Big data refers to large amounts of both structured (numerical data from properly maintained databases) and unstructured (digital pictures and videos, audio recordings and stock ticker data) data. In the early 21st century, Doug Laney, an industry analyst, defined big data using three terms, which later came to be known as the three V’s, namely volume, velocity and variety. Organizations / individuals collect data from several different sources, such as information related to financial transactions, chatter on social media, and even sensor-to-machine or machine-to-machine data. Typically, all of the aforementioned examples represent large volumes of data. Additionally, in most cases, data is usually generated rather rapidly. Therefore, the appropriate technologies, which are capable of handling the speed at which data streams are generated and process them in a timely manner, need to be in place. Finally, there is immense variety in different data streams, which may either be structured or unstructured.
DEEP LEARNING OVERVIEW
The term Deep Learning was coined in 2006 by Geoffrey Hinton to refer to algorithms that enable computers to analyze objects and text in videos and images. Fundamentally, deep learning algorithms are designed to analyze and use large volumes of data to improve the capabilities of machines. Companies, such as Google, Amazon, Facebook, LinkedIn, IBM and Netflix, are already using deep learning algorithms to analyze the activity of a user and then, make customized suggestions and recommendations based on individual preferences. Deep learning majorly focuses on the following area:
- Drug Discovery: In November 2012, a team led by Geoffrey Hinton from the University of Toronto won a competition organized by Merck on automatic drug discovery, by applying deep learning algorithms. The competitors had to go through database entries comprising of more than 30,000 small molecules and predict how each of them were likely to act on 15 different targets. All the small molecules were annotated and had thousands of numerical chemical-property descriptors. The most interesting aspect of the win was that no one in the team had any background in chemistry, biology or life sciences.
- Image Recognition: In 2012, Google announced that its deep learning algorithm could identify humans and / or cats from an image. The algorithm was trained on ~10 million still photographs extracted from YouTube, on a network of around a thousand computers.
- Street Mapping: In 2014, Google claimed to have mapped every single location in France within two hours. They achieved this by feeding street view images into their deep learning algorithm, which was trained to read and recognize street numbers. Baidu (known as the Google of China) is doing something very similar in China.
- Tumor Prognosis: A research team in Boston claimed to have discovered a number of clinically relevant features on tumors that can help doctors in making better prognosis of cancer.
- Patient Survival Rate: A research team at Stanford University developed a deep learning-based system, which, they claim, is better than human pathologists at analyzing tumor samples and predicting the survival rates associated with cancer patients.
Today, in many ways, deep learning algorithms have enabled computers to see, read and write. Therefore, it is safe to say that, given the high accuracy offered by such solutions, deep learning has revolutionized the medical imaging and diagnostics market.
APPLICATIONS OF DEEP LEARNING IN DRUG DISCOVERY
- Drug Target Identification: One of the primary challenges in drug discovery is identifying suitable drug targets. Deep learning models can analyze vast biological datasets, including genomics, proteomics, and metabolomics data, to predict potential drug targets with high accuracy.
- Drug Design and Optimization: Deep learning accelerates the drug design process by generating molecular structures with desired properties. Generative adversarial networks (GANs) and recurrent neural networks (RNNs) are employed to create novel compounds and optimize existing ones.
- Predicting Drug-Drug Interactions: Understanding potential interactions between drugs is crucial to avoid adverse effects. Deep learning models can predict drug-drug interactions by analyzing chemical structures and biological data, reducing the risk of harmful combinations.
- Drug Repurposing: Identifying new therapeutic uses for existing drugs, known as drug repurposing, is another area where deep learning shines. By analyzing comprehensive databases, deep learning models can suggest novel applications for existing medications, saving time and resources.
CHALLENGES IN DEEP LEARNING FOR DRUG DISCOVERY
While deep learning offers immense potential, it also faces several challenges in the pharmaceutical field:
- Data Availability: Access to high-quality, labeled datasets is critical for training deep learning models. In the drug discipline, obtaining such datasets, especially for rare diseases, can be a challenge.
- Interpretability: Interpreting deep learning models can be difficult, especially for complex tasks like drug discovery. Understanding why a model makes a specific prediction is crucial for regulatory approval and trust in the process.
- Ethical Concerns: Using deep learning to predict patient responses to drugs raises ethical questions about data privacy and informed consent. Striking a balance between innovation and ethical considerations is essential.
CONCLUDING REMARKS
Deep learning offers great promise in the healthcare sector, as a result of which, various technology developers are shifting their focus towards this industry. however, unlike the environment observed in the technology industry, where an unsuccessful model can be easily improved in subsequent versions, the models developed for healthcare sectors have to be the best versions of themselves in order to achieve success. This can be attributed to the fact that the healthcare field is highly regulated and hence, demands significant resources and capital for the launch of any new product.
Experts believe that artificial intelligence has already taken its root in the healthcare domain. Intelligent machines backed by relevant deep learning models are already producing significant results across various segments of the industry. However, the deep learning industry is still in its early phases and hence, it is still very early to conclude whether such models can actually replace physicians. Figuratively, the current state of deep learning models in healthcare can be compared to the very first mobile phone, and it is likely to take a few more years for more advanced and dependable technologies to enter the market. It is also worth noting that the success of the current cognitive models is largely dependent on how they are trained, and the type of input data used. Therefore, the adoption of deep learning algorithms demands combined efforts of physicians, doctors and data scientists, in order to develop intelligent and sentient machines which are capable of accurately and effectively providing healthcare solutions.
In a nutshell, deep learning technique has evolved as an excellent computational resource, that has the ability to process a large amount of data using neural networks. The potential applications of the technique in feature extraction, medical imaging, diagnostics, and drug discovery and development have significantly assisted the healthcare and life sciences domain. Deep learning is a rapidly evolving segment of artificial intelligence, and these both are believed to greatly influence the creation of new business models.
Roots Analysis is a global leader in the pharma / biotech market research. Having worked with over 750 clients worldwide, including Fortune 500 companies, start-ups, academia, venture capitalists and strategic investors for more than a decade, we offer a highly analytical / data-driven perspective to a network of over 450,000 senior industry stakeholders looking for credible market insights. All reports provided by us are structured in a way that enables the reader to develop a thorough perspective on the given subject. Apart from writing reports on identified areas, we provide bespoke research / consulting services dedicated to serve our clients in the best possible way.
