Machine Learning in Pharmacotherapeutics

P. Veeresh Babu^* and Obbu Kavya
^*Correspondence: P. Veeresh Babu, Department of Pharmacology, Gokaraju Rangaraju College of Pharmacy, Bachupally, Hyderabad, India, Tel: 9493530528, Email:

Keywords

Machine learning • Artificial intelligence • Algorithms • Diagnosis • Drug discovery • Drug development • Pharmacotherapeutics

Introduction

The advent of "intelligent" robots and technology, known as artificial intelligence, was originally portrayed as a fantasy by philosophers, artists, and science fiction writers; nonetheless, artificial intelligence is now a part of everyday life and is a cornerstone of medicine and research [1]. Machine learning is an artificial intelligence field in which computer aided models are used to identify and learn patterns in high dimensional data in order to create prediction and classification based on the training data. Arthur Samuel working at IBM was the first person to popularise the phrase machine learning in the 1950’s. Machine learning has been developing since that time [2].

Classifications

Machine learning can be classified accordingly as supervised learning, unsupervised learning and reinforcement learning [3]. Models of reinforcement learning are trained using direct reward and punishment as feedback for positive and negative performance [3]. Positive feedback (reward) effectively educates the machine learning model to make the samedecision again in the future, whereas negative feedback (punishment) successfully trains the machine learning model to avoid making the decision in the future comparison to supervised or unsupervised machine learning methods, reinforcement learning plays a very minor part in precision medicine approaches due to the direct feedback required [5].

Objectives

The aim of unsupervised machine learning is to find patterns in unlabeled data as a result it can automatically detect clusters of related cases within a dataset. Once detected, separate clusters are visualised, further studied, analysed for example using standard statistics. Unsupervised machine learning methods can be especially useful in answering questions like "Are there different types of disease?" and other questions regarding diagnosis etc. [6]. Principal component analysis, k-nearest neighbours, and variational autoencoders (an unsupervised deep learning architecture) are examples of popular unsupervised machine learning models [7]. In contrast, supervised machine learning algorithms seek to find patterns in multidimensional data based on labeled data (for example, healthy vs. sick or outcome scores) [8]. A training dataset with ground truth labels is often used to develop a model and maximize machine learning model performance for the desired outcome [9]. The patterns that have been discovered (learned) can subsequently be utilized to classify new datasets or produce datadriven, patient-individual predictions. To categorize datasets, machine learning classification models are utilized, whereas regression models are often used to predict continuous result scores. Support vector machines, random forests, linear models, and deep neural networks are examples of popular supervised statistics and machine learning techniques. In many cases, matching machine learning models (e.g., support vector machines for classification and support vector regression for continuous outcome sources) are available for both issues.

Factors affecting machine learning

Several factors must be considered in order for machine learning techniques to be properly used. First and foremost, the input data must be of good quality, with few artifacts or noise levels. The validity of the ground truth labels is second, and even more critical than a low noise level. While machine learning models can deal with noisy input to some extent, incorrect labels can significantly degrade a machine learning model's performance and are difficult to detect during training. During the data curation process, the correctness of the ground truth labels must be ensured, for example, by having correct diagnosis and diagnostic labels. In general, errors in ground truth labels have a greater detrimental impact on machine learning model performance than other factors. Third, most machine learning models, like many statistical methods, require a training set with no missing characteristics. As a result, it is critical that the training sets be as full as feasible. While data augmentation approaches such as random imputation and more advanced machine learning-based algorithms can be used to fill in the missing data, they rarely produce the same results as using a complete dataset for training. fourth, bigger Datasets are often preferable because they allow the machine learning model to learn the genuine variation in the data with a lower danger of a few outliers negatively affecting the model. However, one of the most significant barriers to developing advanced machine learning models for precision medicine techniques, particularly in the setting of rare diseases, is the collection of large enough datasets. Fifth, precision and accuracy are both crucial in classification methods, and machine learning models should be improved with both in mind. Finally, before applying a machine learning model for computer-aided diagnosis support or clinical treatment decision making, it should be validated. A machine learning model, for example, can be trained using datasets including cases identified by a physician. Following that, new cases can be independently created. Following that, additional instances can be assessed independently by a physician and the machine learning model, and the findings can be compared to validate the model [10].

There are numerous classification and regression models within supervised learning that can be utilized to solve a given problem, each of which has benefits and drawbacks when examining a specific clinical condition [11].

Machine learning using artificial neural networks

The interconnection of neurons and structures observed in the human brain inspired the development of artificial neural networks. Artificial neural networks are made up of an input layer, several hidden layers, and an output layer [12]. Each layer consists of many artificial neurons that are typically connected to the neurons in the following layer via so-called weights, which duplicate the axons and dendrites that connect brain cells (this is known as a feed-forward network). In the case of a classification problem, the number of neurons in the input layer is often equal to the number of input features, but the number of neurons in the output layer is typically equal to the number of classes, such as patients with or without a disease. The network's ability to tackle more difficult non-linear problems grows as the number of hidden layers and neurons in each hidden layer increases. However, the network becomes increasingly difficult to optimize and train as time passes [13]. Deep neural networks are artificial neural networks that have several hidden layers.

Literature Review

History

Artificial neural networks have a long history, dating back to 1943, when Walter Pitts and Warren McCulloch presented the first computer model of a neuron. Alexey Ivakhnenko first described miniature artificial neural networks with up to eight layers of interconnected artificial neurons between 1965 and 1971. However, given the technology and techniques available at the time, training these artificial neural networks was computationally highly expensive, and as a result, the popularity of this machine learning type was limited. The next breakthrough came in the 1980's, when James Hopfield presented the first version of a recurrent neural network Hopfield and Geoffrey Hinton and colleagues Rumelhart et al., popularized back propagation for neural network training, allowing for much more efficient training of neural networks. However, it wasn't until the early 2000's that new hardware, particularly specialized graphical processing units, and even more efficient algorithms became available, allowing for the development and training of artificial neural networks with complex architectures and many hidden layers, which soon began to outperform many "classical" machine learning techniques like support vector machines and random forests.

Many different specialized neural network architectures have been developed, and deep learning machine learning models have lately outperformed classic machine learning models for many issues, even outperforming humans in some tasks. Recurrent neural networks, for example, including long short-term memory deep learning, are used to analyze sequential time-series data, but convolutional neural networks are excellent tools for solving difficult signal and image processing tasks, such as automatic picture-based tumor staging.

Convolutional neural networks use convolutional filters in the first layers of deep neural networks to detect crucial temporal or spatial patterns in datasets such as whole genome data or three-dimensional imaging data. Convolutional filters are optimized in the same way as network weights. Thus, rather than handcrafting feature extraction filters, the raw dataset is used directly as an input to the deep convolutional neural network, and the appropriate temporal or spatial features for the machine learning problem at hand are determined automatically [14].

Deep learning approaches, such as convolutional neural networks, will most certainly play a growing role in precision medicine applications in the future, given their capability and potential.

Applications

While machine learning has demonstrated promising outcomes in pharmacotherapeutics, its applications are still growing, and problems such as data quality, interpretability, and regulatory considerations must be overcome. Nonetheless, machine learning has the potential to revolutionize the field of pharmacotherapeutics by allowing for more efficient and tailored methods to drug research and patient care.

Drug discovery: Machine learning algorithms can identify possible drug candidates by analyzing enormous datasets containing molecular structures, genetic information, and biological experiments. Based on structure-activity connections, binding affinity to target proteins, and other important qualities, these algorithms can forecast the possibility of a chemical becoming a therapeutic candidate. This aids researchers in narrowing their hunt for new medications and speeds up the discovery process [15].

Drug repurposing : Machine learning techniques can be used to find current medications with potential therapeutic benefits on a variety of ailments. Machine learning algorithms can uncover prospective therapeutic candidates for repurposing by evaluating massive datasets of medications, disorders, and their connections. This strategy saves time and resources over traditional medication development because existing pharmaceuticals have already been tested for safety [16].

Personalized medicine: Machine learning enables the development of predictive models that can assess patient data such as genetic information, clinical factors, and therapy responses. These models can assist in identifying patient subgroups that are more likely to respond positively or poorly to a given medicine or treatment. This information enables healthcare providers to customise treatment programs and choose the most appropriate medication for specific patients [17].

Machine learning algorithms for predicting adverse drug reactions: Machine learning algorithms can examine large-scale clinical and genetic information to uncover patterns and connections between medications and adverse reactions. These models can forecast the likelihood of adverse medication reactions by taking into account a variety of factors such as genetic predispositions, drug interactions, and patient characteristics. This information can help healthcare providers make better informed judgments when prescribing pharmaceuticals and aid in drug safety evaluation [18].

Drug dosage optimization: Machine learning algorithms can examine patient-specific parameters such as age, weight, genetics, and biomarker levels to optimize medicine dose. These algorithms can consider a variety of characteristics and provide suggestions for tailored dose regimens that improve efficacy while avoiding side effects or toxicity [19].

Disease diagnosis and prognosis : Machine learning techniques can be used to aid in the diagnosis and prognosis of diseases using medical imaging, electronic health records, and other clinical data. These algorithms can help healthcare workers diagnose diseases, forecast disease development, and choose appropriate treatment options by analyzing patterns and features in patient data [20].

Overall, machine learning plays an important role in pharmacotherapeutics by assisting in drug development, repurposing existing medications, personalizing treatment plans, anticipating adverse drug reactions, optimizing drug dosage, and assisting in illness diagnosis and prognosis. These applications have the potential to increase the effectiveness, safety, and efficiency of medication, resulting in better patient outcomes.

Machine learning plays an important role in the field of pharmacotherapeutics by utilizing computational models to assess enormous volumes of data and aid in drug discovery, development, and individualized treatment. Here's a summary of how machine learning works in pharmacotherapeutics:

Data collection: A large amount of data is gathered, such as clinical trial data, patient records, genomic information, electronic health records, and scientific literature. These data sources reveal important information on illness processes, treatment outcomes, and drug responses.

Data preprocessing: The obtained data frequently needs to be cleansed, standardized, and organized before it can be analyzed further. To ensure consistency and correctness, this stage entails deleting irrelevant or noisy data, addressing missing values, and normalizing the data.

Feature extraction and selection: Relevant features (variables) are extracted from pre-processed data in pharmacotherapeutics. Patient demographics, genetic markers, illness traits, biomarkers, and other important clinical data are examples of these features. To minimize dimensionality and increase model performance, feature selection approaches are used to discover the most useful and predictive characteristics [21].

Model training: Pre-processed and chosen features are used to train machine learning algorithms. Models such as decision trees, Support Vector Machines (SVM), random forests, deep learning models, and others can be used. Labeled data is used to train the models, and the input features are related with known drug reactions, treatment results, or illness states.

Model evaluation and validation: To test prediction accuracy and generalization performance, the trained models are evaluated and validated using distinct datasets. Cross-validation approaches, such as k-fold cross-validation, are widely used to evaluate model performance while avoiding overfitting.

Model deployment: Once the model has proven to be reliable, it can be used in real-world applications. forecasting medication responses, optimizing treatment techniques, discovering potential therapeutic targets, forecasting adverse events, and assisting in drug repurposing are all examples of this.

Continuous learning and refinement: Machine learning models can learn and improve constantly over time. As new data becomes available, the models can be retrained or updated to integrate the most recent information and improve their predicting abilities.

In pharmacotherapeutics, machine learning techniques have several advantages, including the ability to analyze complex and high-dimensional data, identify patterns and associations that humans may miss, facilitate personalized medicine by taking individual patient characteristics into account, and accelerate the drug discovery and development process. However, it is vital to highlight that machine learning models are decision-making tools for healthcare professionals and researchers, and their outputs should always be interpreted. While machine learning has shown enormous promise in pharmacotherapeutics, it is not a silver bullet. To assure drug safety and efficacy, the sector still relies on rigorous experimental design, clinical validation, and regulatory approval. Machine learning techniques, on the other hand, can augment and improve the drug research and development process, resulting in more efficient and targeted therapeutic treatments.

Machine learning, for example, could detect diseases early. Six out of every ten Americans have at least one chronic condition, such as cancer or heart disease. Machine learning can help with cancer diagnosis by detecting, measuring, and analyzing tumors using data from medical images and validated by specialists before being implemented in clinical practice. Machine learning could finish screenings in less time by using its advantage in computer capacity to conduct data and imagery analysis faster than human medical experts can. This could cut referral wait times for high-risk patients while also easing the pressure on clinics dealing with staffing shortages or other issues.

Machine learning technology may potentially increase diagnostic consistency and accuracy by eliminating conditions that contribute to human error. Human specialists, for example, who undertake diagnosis, are affected by factors such as weariness and can differ in their interpretation of data and pictures.

Machine learning could also improve access to health care. In the United States, some regions and populations have restricted access to medical experts. This developing technology has the potential to automate certain operations, reducing clinician workloads and allowing non-specialists to do complex tasks such as cardiac imaging and analysis. This may enable medical personnel to reach a wider percentage of the population through at-home care or smaller clinical settings, allowing more patients to receive care.

How widespread is ML and what could limit its use?

Medical practitioners in the United States use a variety of machine learning tools, with the majority relying on data from imaging such as x-rays or MRIs. Our latest research investigated how machine learning was applied to diagnose five major diseases: Certain malignancies, diabetic retinopathy, Alzheimer's disease, heart disease, and COVID-19. Cancer was the most common current imaging-based application, and machine learning was utilized to detect, quantify, and evaluate tumors and lesions.

While researchers continue to develop AI and machine learning skills in medical diagnostics, these technologies have not been broadly embraced and face a variety of hurdles that prevent them from becoming more commonly used. For example, some medical professionals may be hesitant to deploy machine learning in their clinics until it is proven to be effective. Some medical providers, for example, may be hesitant to deploy machine learning in their clinics until its effectiveness has been widely demonstrated in a variety of therapeutic situations. Some medical practitioners may be unfamiliar with how machine learning might fit into and enhance their workflow, as well as gaps in regulatory advice and regulations, as well as the expense of implementation and maintenance, which may limit its growth and utilization.

Our report outlines these difficulties and suggests policy options for legislators to explore. Policies to encourage or require the evaluation of machine learning diagnostic tools in a variety of realworld scenarios, expand access to high-quality medical data, and foster collaboration among developers, providers, and regulators are examples of these. Check out our reports to learn more about machine learning and AI in medical diagnosis (Figure 1).

Figure 1. Application of machine learning.

At the forefront of ML research in medicine is the diagnosis of diseases and the identification of disorders. More than 800 drugs and vaccines to treat cancer were under clinical trials, according to a 2015 report from pharmaceutical research and manufacturers of America. KnJeff Tyner said that while this is wonderful, it also offers the problem of finding ways to work with all the ensuing data in an interview with Bloomberg Technology. The idea of a biologist collaborating with information scientists and computation lists is crucial in this situation, according to Tyner.

Large companies were among the first to join the movement, especially in sectors with a high unmet need, such the detection and treatment of cancer. In October 2016, IBM Watson Health unveiled IBM Watson genomics, a joint venture with quest diagnostics that combines cognitive computing with genetic tumor sequencing to advance precision medicine.

The Boston-based biopharma business Berg is researching and creating diagnostics and therapeutic treatments, including oncology, utilizing AI. Dosage trials for intravenous tumor treatment, as well as prostate cancer screening and management, are active research initiatives right now.

Other notable instances include Google's DeepMind Health, which last year announced numerous partnerships in the UK, including one with London's Moorfields Eye Hospital, in which they are creating technologies to address macular degeneration in aging eyes.

Predictive analytics are being used by Oxford University's P1vital^® Predicting Response to Depression Treatment (PReDicT) project to aid in the diagnosis and treatment of brain-based illnesses like depression, with the ultimate aim of creating a commercially available battery of emotional tests for use in clinical settings.

Personalized care and behavioral modification

Better illness assessment and personalized medicine, or more effective therapy based on personal health data combined with predictive analytics, are both popular research topics. The dominant technology in the field at the moment is supervised learning, which enables doctors to choose from a smaller number of diagnoses or determine a patient's risk based on their symptoms and genetic information, for example.

In order to choose the best course of therapy, IBM Watson Oncology uses patient medical data and history, making it a prominent institution at the forefront of pushing change in treatment options:

Memorial Sloan Kettering and IBM

A new flood of data will become available during the following ten years thanks to greater use of micro biosensors and devices, as well as mobile apps with more advanced health-measurement and remote monitoring capabilities. This data will be used to support R and D and treatment efficacy. This kind of individualized care has significant consequences for the patient in terms of maximizing health as well as for lowering total healthcare expenses. The reduction in health-care expenses will gradually increase and (hopefully) go back down if more patients follow their doctors' orders and adhere to recommended medications or treatment plans, for example.

In a December interview with Emerj, Catalia Health's Cory Kidd discussed behavioral modification as an important cog in the prevention machine. And there are several start-ups, with varied degrees of success, in the cancer identification, prevention, and treatment arena, for example.

• Somatix, a B2B2C which is adata analytics software platform startup whose ML-based application employs "recognition of hand-to-mouth gestures to help people better understand their behaviour and make life-affirming changes," notably in smoking cessation.
• SkinVision -it is the self-proclaimed "skin cancer risk app" claims to be "the first and only CE certified online assessment." Surprisingly, we were unable to locate SkinVision on the app store. DermCheck was the first app to appear in a "SkinVision" search, in which images submitted based on machine learning. And is also based on based accuracy at scale that still need to be ironed out.

Drug discovery/manufacturing

The use of machine learning in preliminary (early-stage) drug discovery has the potential for a variety of applications, ranging from first drug compound screening to predicting success rate based on biological parameters. This covers research and development discovering technologies such as next-generation sequencing.

Precision medicine, which entails uncovering mechanisms underlying "multifactorial" diseases and, as a result, new therapeutic pathways, appears to be the frontier in this domain. Much of this study involves unsupervised learning, which is still mostly limited to recognizing patterns in data without making predictions (the latter falls under the purview of supervised learning).

The MIT clinical machine learning group is a key participant in this arena, with precision medicine research focused on the development of algorithms to better understand disease processes and design effective treatments for diseases such as type 2 diabetes. In a variety of initiatives, including a collaboration with the Knight Cancer Institute to develop AI technology for cancer precision treatment, with a current focus on developing a method to personalize drug combinations for Acute Myeloid Leukemia (AML) is based on machine learning.

Discussion

Project Hanover at Microsoft

The royal society of the United Kingdom also adds that ML in biomanufacturing for pharmaceuticals is ready for optimization. Data from testing or manufacturing processes has the potential to assist pharmaceutical producers in reducing the time required to make medications, resulting in cheaper prices and enhanced reproducibility.

Clinical trial investigation

Machine learning offers a number of possible applications in shaping and directing clinical trial research. Using advanced predictive analytics to identify clinical trial candidates could draw on a much broader range of data than is currently available, such as social media and doctor visits, as well as genetic information when looking to target specific populations; this would result in smaller, faster, and less expensive trials overall.

For greater safety, ML can also be utilized for remote monitoring and real-time data access, such as monitoring biological and other signals for any sign of harm or death to participants. According to McKinsey, there are numerous other ML applications for increasing clinical trial efficiency, such as determining the best sample size for increased efficiency; addressing and adapting to differences in patient recruitment sites; and using electronic medical records to reduce data errors (for example, duplicate entry).

Radiology and radiotherapy

Dr. Ziad Obermeyer, an assistant professor at Harvard Medical School, claimed in an October 2016 interview with Stat News, "In 20 years, radiologists will not exist in anything resembling their current form." They may resemble cyborgs, with algorithms scanning thousands of papers per minute." In the meantime, Google's DeepMind Health is collaborating with the University College London Hospital (UCLH) to build machine learning algorithms capable of recognizing distinctions in healthy and cancerous tissues in order to optimize radiation treatments.

DeepMind and UCLH are collaborating to use machine learning to accelerate segmentation (while protecting healthy structures) and improve radiation planning accuracy. More on this topic can be found in our article on machine learning in industry.

Intelligent electronic health records

Document classification (for example, sorting patient queries via email) and optical character recognition (converting cursive or other sketched handwriting into digitized characters) are both critical MLbased technologies in advancing the collection and digitization of electronic health information. Two examples of innovations in this field are MATLAB's ML handwriting recognition technology and Google's Cloud Vision API for optical character recognition

Handwritten character recognition using an artificial neural network in MATLAB

The MIT clinical machine learning group is leading the creation of next-generation intelligent electronic health records that will have built-in ML/AI to assist with diagnoses, clinical decisions, and individualized therapy recommendations. According to the research website, MIT states that there is a "need for robust machine learning algorithms that are safe, interpretable, can learn from little labeled training data, understand our language, and generalize well according to medical settings and institutions."

Prediction of epidemic outbreak

Based on data acquired from satellites, historical information on the web, real-time social media updates, and other sources, ML and AI technologies are also being used to monitor and predict epidemic outbreaks around the world. The opioid epidemic is a perfect illustration of how AI technology is being used today. Support vector machines and artificial neural networks, for example, have been used to forecast malaria outbreaks using data such as temperature, average monthly rainfall, total number of positive cases, and other variables.

Predicting outbreak intensity is especially important in third-world nations, which frequently lack medical infrastructure, educational opportunities, and treatment access. ProMED-mail is an internetbased reporting platform that monitors developing illnesses and provides real-time epidemic reports.

Overcoming difficulties

There are significant difficulties to overcome in the rush to deploy ML technologies to pharma and medicine:

• One of the most critical topics to address right now is data stewardship. Medical data is still personal and difficult to acquire, and it appears reasonable to presume that the majority of the public is hesitant to release data due to data privacy concerns. Surprisingly, according to a March 2016 Wellcome Foundation survey on public attitudes on commercial access to health data in the UK, only 17% of respondents would ever consent to their anonymized data being shared with other parties, including for research.
• Streamlining electronic records, which are currently disorganized and dispersed across systems, will be a critical first step in scaling up individualized treatment solutions.

Applications for machine learning and location data in industry

• More transparent algorithms are required to meet severe drug development requirements; people must be able to look through the “black box” and grasp the causal logic underlying computer judgments.
• Recruiting data science talent and developing a solid skills pipeline are critical in the pharmaceutical sector.
• Breaking down “data silos” and encouraging a “data-centric view” (i.e. seeing the value in sharing and integrating data) across sectors is critical to shifting the industry’s mindset toward embracing and seeing value in incremental improvements over time. Historically, pharmaceutical firms have been reticent to make changes or support research programs unless there is an immediate and considerable monetary benefit.

There is some stigma associated with employing machine learning and location data in commercial applications, which is understandable given the risks associated with exploitation of individual privacy. However, if we look beneath the surface of society's daily web of interactions, we can see that the location information economy from GPS to radio signal-based triangulation to geo-tagged images and beyond is now almost ubiquitous, from the moment we track our morning commute to the end-of-day search for healthy and convenient take-out for dinner (Figures 2 and 3).

Figure 2. Shows patients and feature.

Figure 3. Protein ligand complex.

Conclusion

Human resources has been slower to embrace machine learning and artificial intelligence than other industries, including marketing, communications, and even health care. Since machine learning is developed it is used in various sorts of fields. Machine learning now has a great value in human resources that cannot be valued and this happened due to development in algorithms that can forecast employee attrition like deep learning neural networks that are moving toward more transparent reasoning in explaining why a specific result or conclusion was reached.

Conflict of Interest

Obbu Kavya and P. Veeresh Babu doesn’t have conflict of interest with the publication of manuscript.

References

1. Chaturvedula, Ayyappa, Stacie Calad�Thomson, Chao Liu, and Mark Sale, et al. "Artificial intelligence and pharmacometrics: time to embrace, capitalize, and advance?" CPT Pharmacometrics Syst Pharmacol 8 (2019): 440.

   [Crossref] [Google Scholar] [PubMed]

2. Libbrecht, Maxwell W, and William Stafford Noble. "Machine learning applications in genetics and genomics." Nat Rev Genet 16 (2015): 321-332.

   [Crossref] [Google Scholar] [PubMed]

3. Dutta, Maitreyee, C Rama Krishna, Rakesh Kumar, and Mala Kalra, etc. Proceedings of International Conference on IoT Inclusive Life (ICIIL 2019), NITTTR Chandigarh, India. Vol. 116. Springer Nature, 2020.
4. Lloyd, Seth, Masoud Mohseni, and Patrick Rebentrost. "Quantum algorithms for supervised and unsupervised machine learning." arXiv preprint arXiv 1307.0411 (2013).

   [Crossref] [Google Scholar]

5. Ayesha, Shaeela, Muhammad Kashif Hanif, and Ramzan Talib. "Overview and comparative study of dimensionality reduction techniques for high dimensional data." Inf Fusion 59 (2020): 44-58.

   [Crossref] [Google Scholar]

6. Sweilam, Nasser H, AA Tharwat, and NK Abdel Moniem. "Support vector machine for diagnosis cancer disease: A comparative study." Egypt Inform J 11 (2010): 81-92.

   [Crossref] [Google Scholar]

7. Orru, Graziella, William Pettersson-Yeo, Andre F Marquand, and Giuseppe Sartori, et al. "Using support vector machine to identify imaging biomarkers of neurological and psychiatric disease: a critical review." Neurosci Biobehav Rev 36 (2012): 1140-1152.

   [Crossref] [Google Scholar] [PubMed]

8. Zou, James, Mikael Huss, Abubakar Abid, and Pejman Mohammadi, et al. "A primer on deep learning in genomics." Nat Genet 51 (2019): 12-18.

   [Crossref] [Google Scholar] [PubMed]

9. Kononenko, Igor. "Machine learning for medical diagnosis: history, state of the art and perspective." Artif Intell Med 23 (2001): 89-109.

   [Crossref] [Google Scholar] [PubMed]

10. Kuhn, Max, and Kjell Johnson. Applied predictive modeling. Vol. 26. New York: Springer, 2013.

    [Crossref] [Google Scholar]

11. Hopfield, John J. "Neural networks and physical systems with emergent collective computational abilities." Proc Natl Acad Sci USA 79 (1982): 2554-2558.

    [Crossref] [Google Scholar] [PubMed]

12. Hopfield, John J. "Neurons with graded response have collective computational properties like those of two-state neurons." Proc Natl Acad Sci USA 81 (1984): 3088-3092.

    [Crossref] [Google Scholar] [PubMed]

13. Rumelhart, David E, Geoffrey E Hinton, and Ronald J Williams. "Learning representations by back-propagating errors." Nature 323 (1986): 533-536.

    [Crossref] [Google Scholar]

14. MacEachern, Sarah J, and Nils D Forkert. "Machine learning for precision medicine." Genome 64 (2021): 416-425.

    [Crossref] [Google Scholar] [PubMed]

15. Rifaioglu, Ahmet Sureyya, Heval Atas, Maria Jesus Martin, and Rengul Cetin-Atalay, et al. "Recent applications of deep learning and machine intelligence on in silico drug discovery: methods, tools and databases." Brief Bioinform 20 (2019): 1878-1912.

    [Crossref] [Google Scholar] [PubMed]

16. Zhang, Lu, Jianjun Tan, Dan Han, and Hao Zhu, et al. "From machine learning to deep learning: progress in machine intelligence for rational drug discovery." Drug Discov Today 22 (2017): 1680-1685.

    [Crossref] [Google Scholar] [PubMed]

17. Anand, Namrata, and Possu Huang. "Generative modeling for protein structures." Adv Neural Inf Process Syst 31 (2018).

    [Google Scholar]

18. Vamathevan, Jessica, Dominic Clark, Paul Czodrowski, and Ian Dunham, et al. "Applications of machine learning in drug discovery and development." Nat Rev Drug Discov 18 (2019): 463-477.

    [Crossref] [Google Scholar] [PubMed]

19. Bakheet, Tala M, and Andrew J Doig. "Properties and identification of human protein drug targets." Bioinformatics 25 (2009): 451-457.

    [Crossref] [Google Scholar] [PubMed]

20. Wei, Zhi, Kai Wang, Hui-Qi Qu, and Haitao Zhang, et al. "From disease association to risk assessment: an optimistic view from genome-wide association studies on type 1 diabetes." PLoS Genet 5 (2009): e1000678.

    [Crossref] [Google Scholar] [PubMed]

21. Rajkomar, Alvin, Jeffrey Dean, and Isaac Kohane. "Machine learning in medicine." N Engl J Med 380 (2019): 1347-1358.

    [Crossref] [Google Scholar] [PubMed]