Clinical diagnosis is to discover various fixed-point characteristics (can refer to the change of the whole in terms of local changes, this Blawell's fixed-point principle even inspired the equilibrium concept of game theory constructed by Nash), and lay the foundation for the final pattern recognition (for example, the detection of various pathogens can explain the occurrence and development of diseases to a large extent), so it is necessary to collect multi-level data and integrate them with certain statistical models to continuously approach the fixed point. Such as the examination of various markers, pathological sections, etc. We must consider the factors of multi-level coupling, i.e., interaction, in the network. This is a multi-layered game of interests, which needs to consider the benefits and costs of specific measures. Therefore, we need to ideally abstract the process of medical diagnosis into a pattern recognition calculation process, and we need the help of large-scale data collection and machine learning algorithms to extract a meaningful linear independent substrate, which can be compared with more diseases through its selective expression. Theoretically, the more substrates we have, the more accurate the disease we can express, i.e., we can construct ABCDEFGHIJK...... and so on, different degrees of expression can be regarded as finding a specific point in this high-dimensional space, that is, a one-dimensional sequence that we can understand A25B26C42D12E35F59G85H26I6 (letters are objects, numbers are expression levels). Of course, this is based on the assumption that we believe that a particular disease can correspond to a certain region of this high-dimensional space, so we can perform a certain cluster analysis on this basis for pattern recognition. Of course, theoretically this is an NP problem, and it is impossible for an operation of this magnitude to be solved in a limited polynomial time, so there is no accurate solution algorithm, and we can only use some approximate algorithms to find this local optimal solution. Personally, my preferred algorithm is a variety of clustering algorithms for machine learning, but for specific boundary divisions, I hope to be able to combine probabilistic algorithms such as Monte Carlo algorithm, which is a good attempt to simulate annealing, genetic algorithms, neural networks, etc. After all, we know that the number of combinations of such data is growing exponentially, and we have to converge quickly, i.e., filter out the range with a greater probability of having an optimal solution for traversal search. This reduction in the range of calculations bears a certain resemblance to Darwin's theory of natural selection, and we can think of species evolution as a calculation at the level of nature, which would be a big picture. Personally, I think that the doctor's specific diagnosis process is also using a similar algorithm to calculate, of course, in the current process, it is more to take a fixed point recognition mode, so as to summarize a certain module, and compare the sequence with the existing experience, which can carry out a certain pattern recognition, that is, the final diagnosis formation. This can import various smith-waterman algorithms and needleman algorithms and other dynamic programming of these sequence comparisons, as well as our favorite BLAST algorithm, in which the search for seed sequences can be regarded as a fixed point search strategy, and then a larger range of comparison is like Bayesian inference to continuously increase the possible probability that is, the final effect, we can theoretically approximate the reality with a certain accuracy, that is, to make a possible diagnosis. As for treatment, we can use a similar idea, change the object to be measured to the object to be acted, that is, various targets, and then through a certain combination of drugs to achieve the optimal efficacy, according to the enlightenment given to us by the Fourier series: a pair of linear independent substrates can traverse every point of space through their selective expression, that is, linear combinations, we have a certain degree of certainty to say that many diseases can play a certain therapeutic role through a certain combination of drugs, and we are sure that there is such a set of optimal ratios. This kind of information, which can be expressed as a certain sequence, is theoretically inevitable, and we just need to know which drugs are and their relative proportions, which is an optimal solution to solve the problem. Although we don't know which drugs correspond to specific diseases, we can build a certain database to build a rough training set through existing experience, and then use various algorithms to continuously approximate this optimal solution. Therefore, we can still use the previous sequence alignment algorithms to continuously approximate this optimal solution, different drugs correspond to different sequences, and we can finally find the best combination that can have a high sequence matching (according to the conclusion drawn from the scoring matrix), we believe that we can build a high correlation with the specific efficacy, it is considered a measure of efficacy, this quantitative idea is very important to us, which is the basis of all operations. I remembered that I had talked to a guy about the idea of this kind of sequence before, when I was dumbfounded by a question, am I going to sequence every patient? I think that the immature idea is that yes, we need to collect large-scale data, but this kind of sequencing is different from the existing deep sequencing of DNA, not to seek the specific sequence of ACGT, but to supplement the data of various measurements (such as creatinine enzyme concentration, hCG concentration, prostate-specific antigen, etc.) in a certain data format to form a data structure such as sequence, and then carry out various analyses according to specific algorithms on this basis, and the development of bioinformatics in this regard gives us great confidence. As for medical history/family history, symptoms/signs, and imaging/pathology examinations, we can think of different levels of characteristics that can be integrated into the original data model as well. This multi-level information can play a role in improving the posterior probability of Bayesian inference, that is, continuously improving the accuracy of diagnosis. Eventually, we hope to be able to construct a disease spectrum like genomics, which is more practical, and can obtain certain information outputs (possible diseases) with certain information inputs (various measured indicators). Of course, we also have academic pursuits, and we think that we can cite the hidden Markov model to explain, and we need to use various statistical tools to derive a high-dimensional metastasis probability matrix, and the specific disease is a specific selective expression. We believe that the higher dimensions are a good way to explain the lower dimensional situation. That is to say, the interaction of various levels that we have been struggling with before, that is, the competitive game at different levels can be macroscopically represented by simple operations at this high-dimensional level. This idea was inspired by the fundamentals of Newton-Leibniz calculus. Of course, all of the above is. As a medical student, I'm thinking about how to smash a doctor's job all day long. I'm just theoretically exploring the complex process of medical diagnosis based on my current knowledge, according to my habit of idealizing, abstracting, and then modeling, which I think is an elegant extrapolation for now. Of course, without concrete implementation, it is a piece of, and it is always a pipe dream if you don't get this kind of diagnostic platform for a day.

The first point of view about disease is that disease is not an evil foreign substance, such as evil spirits, etc., which I think is a relatively macro abstract description, but it has not kept up with the pace of development of the times. Disease is an equivalent state that coexists with a state of health, and as a different sequence of the network, it only differs in probability, and there is no essential difference. The assumption here is that based on the high-dimensional structure of the network, we first reduce the dimension to a certain probability matrix, and different probabilities have a certain proportion, and then reduce the dimension to be specifically expressed as a specific overall permutation and combination, that is, a sequence, which can build a highly correlated relationship with various states of the human body.

This sequence is theoretically able to construct a certain mapping relationship with various symptoms of various diseases, that is, we hope to be able to construct a certain sequence to explain the possible symptoms of various complex diseases, and then develop possible therapies based on this. We not only need to collect large-scale data and conduct certain cluster analysis to define a boundary to define a specific disease, but also need to know that this is a probabilistic expression process, that is, there are specific sequences that may express specific symptoms at the macroscopic group level and also have a certain probability distribution, we can only make statistical level inferences, although our goal is to convert the inference of the same time section at the group level into the inference of all time at the individual level, but the collapse of this path has great randomness, that is, there may be chaotic effects, and we may not be able to make accurate measurements if we go through all the paths, just like a weather forecast.

My personal opinion is that we can construct a certain correspondence based on the distribution of various cells in the tissue, which is a dimensionality reduction treatment relative to the macroscopic level of the body (such as edema, plaque, tumor, etc.): the expression of specific cell proliferation, apoptosis, inflammatory cell migration, various collagen in the extracellular matrix, and so on. We can observe these changes on the pathological section, and we just extract these statistical features to construct a certain probability matrix of high-dimensional structures (statistics can extract high-dimensional structures, so as to understand the changes at the low-dimensional level from a higher perspective. And I think the doctor's learning process is also building this high-dimensional structure, so that it can quickly recognize patterns in various situations in reality. Therefore, the statistically derived proportion is important to be able to serve as a precursor probability for Bayesian inference to quickly converge the traversal of all diseases into a finite path finding). And this pattern recognition is multi-layered, and we can look for patterns not only at the cellular level, but also at the gene protein expression level. This method of first lowering and then ascending the dimension can enable us to find the immovable point of the final disease level.

Moreover, we note that there is also a certain convergence in the relationship between cells, that is, the proportion of various cells cannot be changed indefinitely, which makes the combination of cells not exponentially explode, which may make our algorithm no longer need to solve the problem of infinite combination of NP difficult. This mechanism, combined with game theory, we believe that it is an equilibrium reached by competitive games between cells, and even between groups formed by cell combinations. At present, due to the lack of knowledge, I can only roughly deal with a certain matrix relationship to limit each other, that is, the proportion between different objects can only be maintained within a certain range. That is, the state proportion of the probability matrix is fine-tuned in a specific part, and the next question is where to adjust how much, which can only rely on statistics.

Therefore, the collection of data is the basis for pattern recognition for various statistical analyses. Doctors can learn to do this kind of data collection, and we need more low-level data, because computers are not yet intelligent enough to understand natural language, and we need to translate into languages that computers can understand, i.e., all kinds of quantitative data. Therefore, we need a good definition and data structure to store this data, and organize it in a certain arithmetic structure, and our consistent idea is to build probabilistic connections between different sequences, which can be expressed as certain custom functions.

Then, based on these underlying data, we can select specific path collapse according to different levels of target sites, so as to be able to fit specific situations with greater precision, such as the relative proportions of various cells, the relative proportions of the expression of different gene proteins, and so on, corresponding to the symptoms of various diseases: peeling, papules, blisters, inflammatory reactions, etc. Then we can carry out a certain treatment at this level, that is, to find various possible therapies, which is also the development direction of individualized precision medicine in the future. What we are looking for is a closed/active patch of Dort's pattern that can play a role at the overall level, and the macro representation is the shift of the state balance towards the offset of the disease state towards the healthy state. The more specific expression is the specific treatment of various objects, such as various receptor antagonists can play a more effective role, such as proton pump inhibitors and H2 receptor antagonists can play a better role in peptic ulcer, which reveals that we can seek to apply treatment to a limited fixed point to produce a better effect, and then continue to optimize, which is also a kind of dynamic programming of the local optimal to the overall optimal approximation. After all, if we seek a perfect match from the beginning, even if we can find the best treatment sequence, the specific drug may not be available and will not be able to exert an effect. This is a kind of nesting of levels, and we are able to exert a certain influence on these sequences by looking for the fixed points of immobility, that is, through the traversal of the levels, to find the various measures that our reality level can exert influence, such as surgery, drugs, and various alternative therapies. For example, if we want to affect the ABCDEFGHIJKLMN sequence (relative to the fixed point of the disease), but there may be a certain combination of effects, that is, we can treat the DFGI (relative to the fixed point of the sequence) to play a similar role, and the latter implementation is more realistic, after all, there are many fewer variables. Let's take an example as proof: suppose there is a disease (too many, just flip through it all), which is a high-dimensional structure, and then we can carry out a certain decomposition and dimensionality reduction, that is, the symptoms are broken down into specific changes at the tissue level and various changes at the cellular level, and even down to the underlying gene expression network, which is the sequence we have been talking about, a basic structure of data storage and computing. For example, the increased expression of collagen fibers in nodules and the absence of lymphocytes, mucin deposition in papules and the distribution of fibroblasts, etc., are all certain information and can be represented in certain sequences. Then we can choose certain measures to make certain changes to these sequences, theoretically the final is to carry out a certain treatment for each whole of the diagnostic sequence, but unfortunately this is difficult to do, we must be realistic, consider a local optimization idea, that is, through the processing experiment of the finite sequence as a whole, find that the drug combination that may have higher efficacy, and then continue to optimize on this basis. Although there is a certain risk that the overall optimal can not be found, but as a compromise of reality, it has a great effect, and you don't see that the development of new drugs can only be marketed with better efficacy than old drugs? This kind of thinking is the basis of the current big bang of science and technology, and everyone is constantly advancing on the basis of their predecessors.

(I can't help but complain: the symptom of wool is not the root cause, this is a misunderstanding, this is essentially a kind of local optimal search to constantly find the final overall optimum, that is, the so-called root cause, which I understand with the concept of immobility.) But the problem is, I can't even treat the symptoms, can you trust me to cure the root cause? I need to keep accumulating my reputation and qualifications to keep getting closer. As soon as I got started, I had to cure the root cause of the problem, I needed to accumulate enough information to continue to converge to the final optimal solution through Bayesian inference, but this requires large-scale calculations, and only after collecting enough data to form a certain database can I achieve fast pattern recognition, just like the old Chinese medicine. In this way, you also need a certain algorithm, that is, the emphasis on various understandings, you are not smart enough to form this kind of algorithm to calculate all kinds of data, that is, diagnosis and treatment. And the power of science lies in allowing the dumbest people to use the tools invented by smart people, and this raising of the lower limit is the embodiment of power, see that the middle and lower classes of people in developed countries generally live better than the middle and upper class people in backward countries. However, it may be that my understanding of TCM is not enough, and it may be able to provide us with better algorithm ideas, which can be better than many previous approximate algorithms, and can have greater efficiency to find the overall optimal result)

But this is the ultimate goal of the future, and we need to make some compromises now, first to prove that this methodology works, and then to import it into the current system of human social collaboration, so that more people can move along this path. Just like the invention of PCR and various hybrid blotting techniques, the study of cell organisms has been very different. (It's a bit delusional again, although young people should dream, but I seem to have too many dreams)

If you see this, you can confidently say that theoretically all diseases must have a certain treatment that can play a certain role in treatment, we just don't know the specific method. Anyway, I was always filled with this kind of self-confidence, and of course, because of the lack of knowledge, I was also full of inferiority, which is why there is such a strange contradiction in my personal character: confidence in the theory of existence and an inferiority in the powerlessness of concrete realization.

I suddenly thought of a saying: when I was studying medicine, I felt that there was no disease in the world and it was incurable, but after practicing medicine, I found that there was no disease in the world that could be cured. Maybe I'm still in the front state, overly optimistic, although I won't enter the back state, but we must clearly understand that this is the different state of mind of human beings for the known and the unknown, and the greatness of human beings lies in the ability to constantly infer the unknown from the known, constantly expanding our cognitive boundaries. Therefore, I have done this philosophical discussion, hoping to get out of this strange circle.