From Population Averages to Network Gaps: The Methodological Shift Behind the SEMO Patent

On May 8, 2026, the China National Intellectual Property Administration granted the invention patent "Method, System and Application for Generating Compound Intervention Schemes Based on Pre-trained Models" (CN117766054B). This patent corresponds to our technical exploration around SEMO: how to put compounds, protein-protein interaction networks, individual omics data, and intervention scheme generation into a computable framework.

I prefer to understand it as a methodological shift:

> From "checking whether an indicator is below the population average" to "examining whether there exists an intervenable gap within the individual's network."

1. Why "Averages" Are Not Enough

Modern health management is largely built on population normal values.

Blood calcium, vitamin D, ferritin, blood sugar, blood lipids, uric acid, hormone levels... most indicators have a reference range. Falling within the range is considered "normal"; falling outside is considered "abnormal."

This approach is important. It allows medical judgments to be standardized and clinical work to be scaled.

But it has a hidden premise:

> It treats a person as a position relative to a population distribution.

In other words, it answers:

Are you higher or lower compared to most people?

But the real question in personalized health is often not this.

The more critical questions are:

- Which modules in this person's body network are under stress?

- Which modules still have reserves?

- Which intervention directions might truly correspond to this person's current network state?

This is the problem we originally wanted to solve when building SEMO.

We are not trying to negate population normal values. Rather, we want to add an internal reference beyond them:

> Let a person's biological network compare with itself.

2. The Body Is Not an Indicator Table, But a Network

After a compound enters the body, it doesn't just change a single isolated indicator.

It may affect certain proteins, trigger a set of interacting gene networks, and further influence metabolism, immunity, inflammation, mitochondrial function, neurotransmitters, DNA repair, or cellular stress responses.

So the real question is not "whether this ingredient is theoretically good" or "whether this indicator is within the normal range," but:

> Do the targets associated with this compound correspond to a real local state difference in this person's protein-protein interaction network?

The core consideration of SEMO is to compute this difference.

We map the known targets of a compound onto the PPI network, and then compare these targets with the non-target background in the same local network. The individual data used here can be DNA methylation, and can be further extended to other omics levels.

If a stable signal difference emerges between the target region and the background region, we can interpret it as a "network gap."

This term is quite evocative.

On a water conservancy map, water level differences represent potential energy and also the direction of flow. In a biological network, the signal gap between a target region and a background region may also represent a state shift worth paying attention to.

!Gene Network Gap Landscape

It is not a single-point indicator, nor a population average, but a relative relationship within an individual's internal network.

3. The Implicit Assumption Behind SEMO

There is a very simple assumption behind SEMO:

> Whether an intervention direction is worth pursuing depends not only on how popular the compound itself is, but on whether the network region it acts on truly presents a computable difference in this individual.

This is quite different from the traditional supplement logic.

The traditional logic is often:

- What is generally deficient in this age group;

- A certain compound has shown effectiveness in research;

- A certain population benefits on average;

- A certain indicator is low, so supplement something.

SEMO wants to do something different:

> Transform "is this thing good or not" into "does the network module corresponding to this thing have an observable gap in this person."

This is also why we emphasize PPI networks.

Disease is not a single gene problem, and aging is not a single indicator problem. A living system is more like a high-dimensional network: different nodes are interconnected, and different modules compensate for each other. A place may appear normal while another place has already developed compensation; an indicator may not seem low, yet the related network may already be under stress.

Therefore, SEMO is not simply doing "ingredient recommendation."

It is more like doing something foundational:

> Turning the relationship between compound action and individual network state into a computable, sortable, and further verifiable structure.

4. Why This Matters

Precision health and personalized medicine have long faced a challenge: we all know individual differences matter, but it's hard to truly transform individual differences into computational structures.

Most of the time, so-called "personalization" still remains at the label level: age, gender, lifestyle, symptoms, questionnaires, and a few test indicators.

This information is useful, but not enough.

True personalization should go deeper, to the network level:

- What state is this person's immune network in?

- Does the metabolic network have local stress?

- Have mitochondria-related modules shown signal shifts?

- Does the target region corresponding to a certain natural product or drug repurposing candidate happen to be in this person's network gap?

If these questions cannot be computed, personalization can only remain at the level of empirical judgment.

The significance of SEMO is that it pushes these questions one step further.

It doesn't directly give ultimate answers, but first turns the questions into a computable form:

> For each person, establish a mapping of "compound × PPI local network × individual omics state."

Once this mapping is established, it becomes possible to do ranking, tracking, retesting, verification, and iteration.

5. From "What Am I Missing" to "Where Are the Gaps"

I've always felt that the question "what am I deficient in" has been asked too simplistically in the health industry.

Many times, when people ask "what am I missing," they're actually asking:

What should I take?

But the body may not answer questions this way.

What the body truly presents to us may not be a simple deficiency checklist, but a complex network state diagram.

- Some areas may appear to have normal indicators, yet the network is already under stress;

- Some areas may have slightly low indicators, yet the system can still maintain stability;

- Some compounds are theoretically excellent, but don't correspond to this person's main network gaps;

- Some less popular compounds might actually hit an overlooked local module.

Therefore, we want to rewrite the question:

> Not "what am I deficient in," but "where do gaps appear in my network."

Going further:

> Not "which compound is best," but "which compound's corresponding network region is most worth verifying in my body."

This is the real consideration behind SEMO.

It's not about creating more supplementation demand, but about reducing blind experimentation.

6. Why Pre-trained Models Are Needed

If we only look at one compound, one network, and one sample, the problem isn't very complex.

But in the real world, there are many compounds, many PPI subnetworks, and many individual samples. Each compound may correspond to multiple targets, and each target is situated in different network backgrounds.

This creates an enormous computational space.

In our bioRxiv preprint, we proposed the idea of using pre-trained network medicine models to generate biomarkers, targets, drug repurposing leads, and personalized schemes. The core approach is to decompose the human PPI network into a large number of subnetworks, and then combine them with numerous compound targets to form a reusable SEMO feature pool.

This step is crucial.

Because if personalized medicine has to compute from scratch every time, it's difficult to scale. The significance of pre-trained models lies in:

> First structuring the possible "compound-network" relationships, then mapping them onto specific individual data.

In other words, SEMO is not a single analysis, but a foundational representation system.

It first establishes a large-scale network medicine feature space, and then lets each individual's omics data enter this space to find the network gap profile belonging to that person.

7. Relationship with the FDA's Trend Toward Mechanistic Evidence

In recent years, the FDA's discussions on Plausible Mechanism indicate an important trend: in highly individualized scenarios where traditional large-scale clinical trials are difficult to fully conduct, regulatory science is placing greater emphasis on the combination of mechanistic evidence, natural history, and limited clinical data.

This is a signal for the entire precision medicine field.

Future medical evidence will not only speak the language of population average effects. We need another language:

- Why does this intervention make sense for this person?

- Which molecular abnormality, which network module, which mechanistic chain does it act on?

- Can this mechanistic change be observed and tracked?

What SEMO does has an inherent resonance with this trend.

It attempts to interpret compound action within individual networks, transform intervention directions from empirical judgment into mechanistic hypotheses, and advance personalized schemes from "recommendation lists" to "network evidence structures."

This is also why I believe SEMO will be truly important in the future.

It is not an isolated algorithm, but part of the infrastructure for individualized mechanistic evidence.

8. The Significance of the Patent Grant

The invention patent granted this time is titled "Method, System and Application for Generating Compound Intervention Schemes Based on Pre-trained Models."

What it covers is not just a single-point calculation formula, but an entire technical chain:

- Organization of compound target information;

- Construction of PPI networks and subnetworks;

- Extraction of individual omics features;

- Generation of compound-related network features;

- Ranking and output of intervention schemes based on pre-trained models.

From an engineering perspective, this means SEMO is no longer just a research idea, but a technical framework that can enter system implementation, product transformation, and continuous iteration.

From a scientific perspective, it also raises a new question:

> Can we use network gaps, rather than single-point indicators, to reorganize the evidence for personalized interventions?

This may be the deeper significance behind this patent.

9. What I Hope SEMO Will Ultimately Become

I don't want SEMO to be just a "supplement recommendation" tool.

If it only tells users what to take, that would be too shallow.

I hope it will ultimately become a life network interpretation tool:

- Helping us understand a person's systemic state;

- Helping us discover local network stress that traditional indicators cannot see;

- Helping us connect compound action with individual differences;

- Helping us design better N-of-1 verifications;

- Helping us advance health interventions from empirical consumption to mechanistic science.

The question SEMO truly wants to answer is not "which ingredient is trendiest," but:

> In this person's life network, where do there exist gaps that can be understood, tracked, and verified?

If this is accomplished, precision health will no longer be just a marketing term.

It will become a personal science that can be repeatedly computed, retested, and iterated.

10. Conclusion: Let the Body's Network Speak for Itself

In the past, we've been too accustomed to interpreting the body through external standards.

Population averages, normal ranges, epidemiological conclusions, expert experience, product recommendations... these are all important, but none of them are this person's own network.

SEMO wants to do something different:

> Let the individual's molecular network become the starting point for judging intervention directions.

Moving from averages to network gaps is not about abandoning traditional medicine, but about adding a perspective closer to the system itself beyond traditional indicators.

When we can see the network gaps inside a person's body, we have the opportunity to advance health management from "guessing what I should supplement" to "verifying where my network needs to be remodeled."

This is the implicit consideration behind SEMO.

And the direction this patent truly wants to open up.

References

1. Xiong J. Method, System and Application for Generating Compound Intervention Schemes Based on Pre-trained Models. Invention Patent, Grant Number: CN117766054B, Grant Date: 2026-05-08. Applicant: Beijing Deep Methylation Health Technology Co., Ltd.

2. Xiong J. Utilizing Pre-trained Network Medicine Models for Generating Biomarkers, Targets, Re-purposing Drugs, and Personalized Therapeutic Regimes: COVID-19 Applications. bioRxiv. 2023. doi: 10.1101/2023.02.21.527754.

3. U.S. Food and Drug Administration. Considerations for the Use of the Plausible Mechanism Framework to Develop Individualized Therapies that Target Specific Genetic Conditions with Known Biological Cause. Draft Guidance for Industry. February 2026.

4. Barabási AL, Gulbahce N, Loscalzo J. Network medicine: a network-based approach to human disease. Nature Reviews Genetics. 2011;12:56–68.