Glass Box and Black Box Machine Learning Approaches to Exploit Compositional Descriptors of Molecules in Drug Discovery and Aid the Medicinal Chemist.

The synthetic medicinal chemist plays a vital role in drug discovery. Today there are AI tools to guide next syntheses, but many are "Black Boxes" (BB). One learns little more than the prediction made. There are now also AI methods emphasizing visibility and "explainability" (thus explainable AI or XAI) that could help when "compositional data" are used, but they often still start from seemingly arbitrary learned weights and lack familiar probabilistic measures based on observation and counting from the outset. If probabilistic methods were used in a complementary way with BB methods and demonstrated comparable predictive power, they would provide guidelines about what groups to include and avoid in next syntheses and quantify the relationships in probabilistic terms. These points are demonstrated by blind test comparison of two main types of BB methods and a probabilistic "Glass Box" (GB) method new outside of medicine, but which appears well suited to the above. Because many probabilities can be involved, emphasis is on the predictive power of its simplest explanatory models. There are usually more inactive compounds by orders of magnitude, often a problem for machine learning methods. However, the approaches used here appear to work well for such "real world data".

Principles of Quantum Mechanics for Artificial Intelligence in medicine. Discussion with reference to the Quantum Universal Exchange Language (Q-UEL).

This paper reviews some basic principles of Quantum Mechanics, Quantum Computing, and Artificial Intelligence in terms of a specific unifying theme. This theme relates to the hyperbolic or split-complex imaginary numbers and their equivalent matrices, rediscovered by Dirac, and the underlying mathematics of the previously described Q-UEL language based on them. Hyperbolic imaginary numbers h have the property hh = +1: contrast the more familiar i such that ii = -1. Examples of analogous matrices include that for the Hadamard gate as used in quantum computing and the Pauli spin matrices, and all Hermitian matrices of interest in quantum computing can readily be derived from these. They also relate to Dirac dualization, spinor projectors of Quantum Field Theory, the non-wave-like part of quantum theory, collapse of the wave function, and a dualized form of classical probability theory that has advantages in automated reasoning for medicine.

De novo protein folding on computers. Benefits and challenges.

There has been recent success in prediction of the three-dimensional folded native structures of proteins, most famously by the AlphaFold Algorithm running on Google's/Alphabet's DeepMind computer. However, this largely involves machine learning of protein structures and is not a de novo protein structure prediction method for predicting three-dimensional structures from amino acid residue sequences. A de novo approach would be based almost entirely on general principles of energy and entropy that govern protein folding energetics, and importantly do so without the use of the amino acid sequences and structural features of other proteins. Most consider that problem as still unsolved even though it has occupied leading scientists for decades. Many consider that it remains one of the major outstanding issues in modern science. There is crucial continuing help from experimental findings on protein unfolding and refolding in the laboratory, but only to a limited extent because many researchers consider that the speed by which real proteins folds themselves, often from milliseconds to minutes, is itself still not fully understood. This is unfortunate, because a practical solution to the problem would probably have a major effect on personalized medicine, the pharmaceutical industry, biotechnology, and nanotechnology, including for example "smaller" tasks such as better modeling of flexible "unfolded" regions of the SARS-COV-2 spike glycoprotein when interacting with its cell receptor, antibodies, and therapeutic agents. Some important ideas from earlier studies are given before moving on to lessons from periodic and aperiodic crystals, and a possible role for quantum phenomena. The conclusion is that better computation of entropy should be the priority, though that is presented guardedly.

Mining real-world high dimensional structured data in medicine and its use in decision support. Some different perspectives on unknowns, interdependency, and distinguishability.

There are many difficulties in extracting and using knowledge for medical analytic and predictive purposes from Real-World Data, even when the data is already well structured in the manner of a large spreadsheet. Preparative curation and standardization or "normalization" of such data involves a variety of chores but underlying them is an interrelated set of fundamental problems that can in part be dealt with automatically during the datamining and inference processes. These fundamental problems are reviewed here and illustrated and investigated with examples. They concern the treatment of unknowns, the need to avoid independency assumptions, and the appearance of entries that may not be fully distinguished from each other. Unknowns include errors detected as implausible (e.g., out of range) values that are subsequently converted to unknowns. These problems are further impacted by high dimensionality and problems of sparse data that inevitably arise from high-dimensional datamining even if the data is extensive. All these considerations are different aspects of incomplete information, though they also relate to problems that arise if care is not taken to avoid or ameliorate consequences of including the same information twice or more, or if misleading or inconsistent information is combined. This paper addresses these aspects from a slightly different perspective using the Q-UEL language and inference methods based on it by borrowing some ideas from the mathematics of quantum mechanics and information theory. It takes the view that detection and correction of probabilistic elements of knowledge subsequently used in inference need only involve testing and correction so that they satisfy certain extended notions of coherence between probabilities. This is by no means the only possible view, and it is explored here and later compared with a related notion of consistency.

Computers and preventative diagnosis. A survey with bioinformatics examples of mitochondrial small open reading frame peptides as portents of a new generation of powerful biomarkers.

The present brief survey is to alert developers in datamining, machine learning, inference methods, and other approaches in relation to diagnostic, predictive, and risk assessment medicine about a relatively new class of bioactive messaging peptides in which there is escalating interest. They provide patterns of communication and cross-chatter about states of health and disease within and, importantly, between cells (they also appear extracellularly in biological fluids). This chatter needs to be analyzed somewhat in the manner of the decryption of the Enigma code in the Second World War. It could lead not only to improved diagnosis but to predictive diagnosis, prediction of organ failure, and preventative medicine. This involves peptide products of short reading frames that have been previously somewhat neglected as unlikely gene products, with probably many examples in nuclear DNA, but certainly several known in the mitochondrial DNA. There is a great deal of knowledge now becoming available about the latter and itis believed thatthat the mRNA can be translated both by standard cytosolic and mitochondrial genetic codes, resulting in different peptides, adding a further level of complexity to the applications of bioinformatics and computational biology but a higher level of detail and sophistication to preventative diagnosis. The code to crack could be sophisticated and combinatorically complex to analyze by computers. Mitochondria may have combined with proto-eucaryotic cells some 2 billion years ago, only about a 7th of the age of the universe. Cells appeared some 2 billion years before that, also with possible signaling based on similar ideas. This makes life small in space but huge in time, refinement of which centrally involves these signaling processes.

Testing machine learning techniques for general application by using protein secondary structure prediction. A brief survey with studies of pitfalls and benefits using a simple progressive learning approach.

Many researchers have recently used the prediction of protein secondary structure (local conformational states of amino acid residues) to test advances in predictive and machine learning technology such as Neural Net Deep Learning. Protein secondary structure prediction continues to be a helpful tool in research in biomedicine and the life sciences, but it is also extremely enticing for testing predictive methods such as neural nets that are intended for different or more general purposes. A complication is highlighted here for researchers testing their methods for other applications. Modern protein databases inevitably contain important clues to the answer, so-called "strong buried clues", though often obscurely; they are hard to avoid. This is because most proteins or parts of proteins in a modern protein data base are related to others by biological evolution. For researchers developing machine learning and predictive methods, this can overstate and so confuse understanding of the true quality of a predictive method. However, for researchers using the algorithms as tools, understanding strong buried clues is of great value, because they need to make maximum use of all information available. A simple method related to the GOR methods but with some features of neural nets in the sense of progressive learning of large numbers of weights, is used to explore this. It can acquire tens of millions and hence gigabytes of weights, but they are learned stably by exhaustive sampling. The significance of the findings is discussed in the light of promising recent results from AlphaFold using Google's DeepMind.

Extension of the Quantum Universal Exchange Language to precision medicine and drug lead discovery. Preliminary example studies using the mitochondrial genome.

The Quantum Universal Exchange Language (Q-UEL) based on Dirac notation and algebra from quantum mechanics, along with its associated data mining and Hyperbolic Dirac Net (HDN) for probabilistic inference, has proven to be a useful architectural principle for knowledge management, analysis and prediction systems in medicine. It has been described in several papers; here is described its extension to clinical genomics and precision medicine. Two use cases are studied: (a) bioinformatics in clinical decision support especially for risk for type 2 diabetes using mitochondrial patient DNA sequences, and (b) bioinformatics and computational biology (conformational) research examples related to drug discovery involving the recently discovered class of mitochondrial derived peptides (MDPs). MDPs were surprising when first discovered as coded in small open reading frames (sORFs), and are emerging as having a fundamental role in metabolic control, longevity and disease. This project originally represented a language specification study relating to what information related to genomics is essential or useful to carry, and what processing will be needed. However, novel aspects introduced or discovered include the HDN-like neural nets and their use, along with more established methods, for prediction of type 2 diabetes, and in particular for proposals for over 80 natural MDPs most of which that have not previously been described at the time of the study, as potential drug lead targets. Also, use of many medical records with simulated joining of mtDNA as performance tests led to some insightful observations regarding the behavior of HDN predictions where independent factors are involved.

Studies in the use of data mining, prediction algorithms, and a universal exchange and inference language in the analysis of socioeconomic health data.

While clinical and biomedical information in digital form has been escalating, it is socioeconomic factors that are important determinants of health on the national and global scale. We show how collective use of data mining and prediction algorithms to analyze socioeconomic population health data can stand beside classical correlation analysis in routine data analysis. The underlying theoretical basis is the Dirac notation and algebra that is a scientific standard but unusual outside of the physical sciences, combined with a theory of expected information first developed for analyzing sparse data but still largely confined to bioinformatics. The latter was important here because the records analyzed (which are for US counties and equivalents, not patients) are very few by contemporary data mining standards. The approach is very unlikely to be familiar to socioeconomic researchers, so the theory and the advantages of our inference nets over the Bayes Net are reviewed here, mostly using socioeconomic examples. While our expertise and focus is in regard to novel analytical methods rather than socioeconomics per se, a significant negative (countertrending) relationship between population health and equity was initially surprising, at least to the present authors. This encouraged deeper exploration including that of the relationship between our data mining methods and traditional Pearson's correlation. The latter is susceptible to giving wrong conclusions if a phenomenon called Simpson's paradox applies, so this is also investigated. Also discussed is that, even for very few records, associative data mining can still demand significant computational resources due to a combinatorial explosion.

Bidirectional General Graphs for inference. Principles and implications for medicine.

Probabilistic inference methods require a more general and realistic description of the world as a Bidirectional General Graph (BGG). While in its original form the Bayes Net (BN) has been promoted as a predictive tool, it is more immediately a way of testing a hypothesis or model about interactions in a system usually considered on a causal basis. Once established, the model can be used in a predictive way, but the problem here is that for a traditional BN the hypotheses or models that can be formed are limited to the Directed Acyclic Graph (DAG) by definition. Three interrelated features are highlighted that represent deficiencies of the DAG which are corrected by conversion to a method based on a BGG: (i) lack of intrinsic representation of coherence by Bayes' rule, (ii) relatedly the need to consider interdependence in parent nodes, and (iii) the need for management of a property called recurrence. These deficiencies can represent large errors in absolute estimates of probabilities, and while relative and renormalized probabilities ameliorate that, they can often make much of a net superfluous through cancelations by division. The Hyperbolic Dirac Net (HDN) based on Dirac's quantum mechanics is a solution that led naturally to avoiding these deficiencies. It encodes bidirectional probabilities in an h-complex value rediscovered by Dirac, i.e. with the imaginary number h such that hh = +1. Properties of the HDN described previously are reviewed (though emphasis is on descriptions in familiar probability terms), the issue of recurrence is introduced, methods of construction are simplified, and the severity of the quantitative differences between BNs and analogous HDNs are exemplified. There is also discussion of how results compare with other approaches in practice.

Studies in using a universal exchange and inference language for evidence based medicine. Semi-automated learning and reasoning for PICO methodology, systematic review, and environmental epidemiology.

The Q-UEL language of XML-like tags and the associated software applications are providing a valuable toolkit for Evidence Based Medicine (EBM). In this paper the already existing applications, data bases, and tags are brought together with new ones. The particular Q-UEL embodiment used here is the BioIngine. The main challenge is one of bringing together the methods of symbolic reasoning and calculative probabilistic inference that underlie EBM and medical decision making. Some space is taken to review this background. The unification is greatly facilitated by Q-UEL's roots in the notation and algebra of Dirac, and by extending Q-UEL into the Wolfram programming environment. Further, the overall problem of integration is also a relatively simple one because of the nature of Q-UEL as a language for interoperability in healthcare and biomedicine, while the notion of workflow is facilitated because of the EBM best practice known as PICO. What remains difficult is achieving a high degree of overall automation because of a well-known difficulty in capturing human expertise in computers: the Feigenbaum bottleneck.

Data-mining to build a knowledge representation store for clinical decision support. Studies on curation and validation based on machine performance in multiple choice medical licensing examinations.

Extracting medical knowledge by structured data mining of many medical records and from unstructured data mining of natural language source text on the Internet will become increasingly important for clinical decision support. Output from these sources can be transformed into large numbers of elements of knowledge in a Knowledge Representation Store (KRS), here using the notation and to some extent the algebraic principles of the Q-UEL Web-based universal exchange and inference language described previously, rooted in Dirac notation from quantum mechanics and linguistic theory. In a KRS, semantic structures or statements about the world of interest to medicine are analogous to natural language sentences seen as formed from noun phrases separated by verbs, prepositions and other descriptions of relationships. A convenient method of testing and better curating these elements of knowledge is by having the computer use them to take the test of a multiple choice medical licensing examination. It is a venture which perhaps tells us almost as much about the reasoning of students and examiners as it does about the requirements for Artificial Intelligence as employed in clinical decision making. It emphasizes the role of context and of contextual probabilities as opposed to the more familiar intrinsic probabilities, and of a preliminary form of logic that we call presyllogistic reasoning.

Studies of the role of a smart web for precision medicine supported by biobanking.

Both the extraction of medical knowledge from data mining many patient records and from authoritative natural language text on the Internet are important for clinical decision support and biomedical research. The samples in biobanks represent a further kind of information repository of recognized increasing importance, so mechanisms being developed for a smart web for medicine should take them into account. While this paper is primarily a review of Quantum Universal Exchange Language as an XML extension to enable a future smart web for healthcare and biomedicine, it is the first time that we have discussed the connection with biobanks and the design of Quantum Universal Exchange Language's XML-like tags to support their use.

Implementation of a web based universal exchange and inference language for medicine: Sparse data, probabilities and inference in data mining of clinical data repositories.

We extend Q-UEL, our universal exchange language for interoperability and inference in healthcare and biomedicine, to the more traditional fields of public health surveys. These are the type associated with screening, epidemiological and cross-sectional studies, and cohort studies in some cases similar to clinical trials. There is the challenge that there is some degree of split between frequentist notions of probability as (a) classical measures based only on the idea of counting and proportion and on classical biostatistics as used in the above conservative disciplines, and (b) more subjectivist notions of uncertainty, belief, reliability, or confidence often used in automated inference and decision support systems. Samples in the above kind of public health survey are typically small compared with our earlier "Big Data" mining efforts. An issue addressed here is how much impact on decisions should sparse data have. We describe a new Q-UEL compatible toolkit including a data analytics application DiracMiner that also delivers more standard biostatistical results, DiracBuilder that uses its output to build Hyperbolic Dirac Nets (HDN) for decision support, and HDNcoherer that ensures that probabilities are mutually consistent. Use is exemplified by participating in a real word health-screening project, and also by deployment in a industrial platform called the BioIngine, a cognitive computing platform for health management.

Suggestions for a web based universal exchange and inference language for medicine. Continuity of patient care with PCAST disaggregation.

We describe here the applications of our recently proposed Q-UEL language to continuity of patient care between physicians, specialists and institutions as mediated via the Internet, giving examples derived from HL7 CDA and VistA of particular interest to workflow. Particular attention is given to the Universal Exchange Language for healthcare as requested by the US President׳s Council of Advisors on Science and Technology (PCAST) released in December 2010, especially in regard to disaggregation of the patient record on the Internet. To illustrate many features and options, one of our most elaborate configurations combining them, for disaggregation and reaggregation, is described. The Q-UEL tags used do not physically join, but query each other from a random mix via the application. Despite the computationally demanding complexity of the configuration with two joining tags for each data tag and four independently evolving keys, plus a valuable but rate limiting isomorphism test, packets of essential clinical data for patient could be recovered and displayed every 2 s for a "club" of 30,000-50,000 patients in the mix. All computation here is on a standard laptop, but for practical use of the Internet to display downloaded data, the above is adequate, so focus is primarily on increasing club size. In practice, it is not necessary that a club comprise an entire nation. Assuming that one does not use purely random assignments of patients to arbitrary clubs, there could for example be a club comprising all schoolchildren in Scotland, or a club comprising all military veterans in Illinois. In such cases, one is typically dealing with clubs each of the order of a mere million patients. Using such club sizes efficiently, and in principle even a club the size of a whole country, appears to be possible.

POPPER, a simple programming language for probabilistic semantic inference in medicine.

Our previous reports described the use of the Hyperbolic Dirac Net (HDN) as a method for probabilistic inference from medical data, and a proposed probabilistic medical Semantic Web (SW) language Q-UEL to provide that data. Rather like a traditional Bayes Net, that HDN provided estimates of joint and conditional probabilities, and was static, with no need for evolution due to "reasoning". Use of the SW will require, however, (a) at least the semantic triple with more elaborate relations than conditional ones, as seen in use of most verbs and prepositions, and (b) rules for logical, grammatical, and definitional manipulation that can generate changes in the inference net. Here is described the simple POPPER language for medical inference. It can be automatically written by Q-UEL, or by hand. Based on studies with our medical students, it is believed that a tool like this may help in medical education and that a physician unfamiliar with SW science can understand it. It is here used to explore the considerable challenges of assigning probabilities, and not least what the meaning and utility of inference net evolution would be for a physician.

Hyperbolic Dirac Nets for medical decision support. Theory, methods, and comparison with Bayes Nets.

We recently introduced the concept of a Hyperbolic Dirac Net (HDN) for medical inference on the grounds that, while the traditional Bayes Net (BN) is popular in medicine, it is not suited to that domain: there are many interdependencies such that any "node" can be ultimately conditional upon itself. A traditional BN is a directed acyclic graph by definition, while the HDN is a bidirectional general graph closer to a diffuse "field" of influence. Cycles require bidirectionality; the HDN uses a particular type of imaginary number from Dirac׳s quantum mechanics to encode it. Comparison with the BN is made alongside a set of recipes for converting a given BN to an HDN, also adding cycles that do not usually require reiterative methods. This conversion is called the P-method. Conversion to cycles can sometimes be difficult, but more troubling was that the original BN had probabilities needing adjustment to satisfy realism alongside the important property called "coherence". The more general and simpler K-method, not dependent on the BN, is usually (but not necessarily) derived by data mining, and is therefore also introduced. As discussed, BN developments may converge to an HDN-like concept, so it is reasonable to consider the HDN as a BN extension.

Suggestions for a Web based universal exchange and inference language for medicine.

Mining biomedical and pharmaceutical data generates huge numbers of interacting probabilistic statements for inference, which can be supported by mining Web text sources. This latter can also be probabilistic, in a sense described in this report. However, the diversity of tools for probabilistic inference is troublesome, suggesting a need for a unifying best practice. Physicists often claim that quantum mechanics is the universal best practice for probabilistic reasoning. We discuss how the Dirac notation and algebra suggest the form and algebraic and semantic meaning of XML-like Web tags for a clinical and biomedical universal exchange language formulated to make sense directly to the eye of the physician and biomedical researcher.

A universal exchange language for healthcare.

We have defined a Universal Exchange Language (UEL) for healthcare that takes a green field approach to the development of a novel "XML-like" language. We consider here what given a free hand might mean: a UEL that incorporates an advanced mathematical foundation that uses Dirac's notation and algebra. For consented and public information, it allows probabilistic inference from UEL semantic web triplet tags. But also it is possible to use similar thinking to maximize the security and analytic characteristics of private health data by disaggregating or "shredding" it. Both are scalable to millions of records that could be spread across the Internet.

Drug discovery using very large numbers of patents: general strategy with extensive use of match and edit operations.

A patent data base of 6.7 million compounds generated by a very high performance computer (Blue Gene) requires new techniques for exploitation when extensive use of chemical similarity is involved. Such exploitation includes the taxonomic classification of chemical themes, and data mining to assess mutual information between themes and companies. Importantly, we also launch candidates that evolve by "natural selection" as failure of partial match against the patent data base and their ability to bind to the protein target appropriately, by simulation on Blue Gene. An unusual feature of our method is that algorithms and workflows rely on dynamic interaction between match-and-edit instructions, which in practice are regular expressions. Similarity testing by these uses SMILES strings and, less frequently, graph or connectivity representations. Examining how this performs in high throughput, we note that chemical similarity and novelty are human concepts that largely have meaning by utility in specific contexts. For some purposes, mutual information involving chemical themes might be a better concept.

Towards intelligent Internet-roaming agents for mining and inference from medical data.

This paper is effectively subtitled "Considerations of Requirements for Programmable Laws of Probabilistic Higher Order Logical Thought". Why such a need? Issues such as privacy, security, bandwidth, and computational power demand not a central analyzing agency, but roaming agents to analyze the global explosion of medical data in many hundreds of petabytes distributed across many sites. They will send back only the conclusions, not the source data. But how will they reach those conclusions? This future pressing need will driving workers to consider Best Practice in inference. Right now, there are diverse approaches to inference, and it is not clear how to unify them into a self-consistent system. For example, there is not even universal agreement on how to treat probabilistic higher order logic. Quantum mechanics is held by many to be a universal system, but produces bizarre predictions for the everyday world of human experience. However, by rotation of the imaginary number i = square root of (-1) quantum mechanics to the split complex number h such that hh = + 1, quantum mechanics becomes an inference system for higher order probabilistic logic. And the system has interesting emergent properties which may shed light on the nature of thought.