Peptides To Proteomics Revolution

KJ Dhaliwal is a serial entrepreneur building Lotus Health AI, a consumer health app. He previously started & sold the Dil Mil app.

getty

Everyone's talking about peptides.

I'd know. I founded Lotus AI to help people understand their health better, and our clinicians now prescribe peptides for thousands of patients, primarily for metabolic, hormonal and endocrine health. That work has given us a front-row seat.

The real story is that we're approaching the ability to read, interpret and eventually intervene in the human body's operating system. Specifically, that operating system is the proteome.

Your Body Runs On Proteins

Every function in your body, from immune response to tissue repair to neural signaling, is executed by proteins. Your DNA is the blueprint. But proteins are the workers, the machinery and the infrastructure.

Here's the pipeline: Cells transcribe DNA into mRNA, which is then translated into chains of amino acids called peptides and polypeptides. These polypeptides then fold into complex three-dimensional structures called proteins, driven by a combination of hydrophobic interactions, hydrogen bonds and electrostatic forces between atoms along the chain. The final shape of a protein determines what it does and what it can interact with.

The sum of all protein interactions in your body is your proteome. It's the real-time execution layer of your biology.

We're Learning To Read The Source Code

Until recently, we could sequence DNA but couldn't reliably predict what protein structures it would produce. That changed in November 2020 when Google DeepMind announced that AlphaFold had essentially solved the 50-year-old protein folding problem. We can now computationally predict many protein structures from amino acid sequences, a breakthrough that won the 2024 Nobel Prize in Chemistry.

Take collagen. It can be written as a string of letters, each representing an amino acid. From that string, we can now predict its three-dimensional structure. That's like going from reading sheet music to hearing the symphony.

AlphaFold isn't perfect. It predicts static structures well for rigid, well-characterized proteins but struggles with conformational dynamics, protein complexes and proteins that switch between multiple folds. It's a massive leap forward, not a complete map. But it opened the door.

And it gets deeper. After a protein is built, cells modify it. They attach phosphate groups (phosphorylation), sugar molecules (glycosylation) and other chemical modifications at specific positions. Each modification changes the protein's shape, which changes its function. These modified versions are called proteoforms.​

Glycosylation, for instance, plays a significant role in how cells communicate. Cell-surface glycans influence everything from receptor signaling and immune recognition to adhesion and trafficking. Researchers have known for over 30 years that glycan diversity encodes biological information, and we're only now developing the tools to read it systematically.

These modifications are specific, positional and increasingly mappable. They follow patterns. And patterns can be learned by machines.

Disease Has A Proteomic Dimension

Many diseases involve dysfunction at the proteomic level: misfolded proteins, dysfunctional modifications, broken signaling cascades. These aren't random. They're systematic failures in proteomic logic, and they often begin long before symptoms appear.

This doesn't mean the proteome is the sole root cause of all disease. Genetics, environment, the immune system, the microbiome and systems-level interactions all matter. But the proteome is where much of this dysfunction manifests and where it can be measured.

This reframes a fundamental problem in medicine. Instead of waiting for symptoms and then reverse-engineering causes, what if we could monitor the proteomic state continuously and catch dysfunction earlier?

I see this play out in our patient population every week. People come to Lotus looking for peptides to address metabolic or hormonal issues, but the underlying story is often proteomic.

The Tools Are Converging

Three things are coming together at once.

First, proteomic profiling is getting dramatically more powerful. Nautilus Biotechnology's Iterative Mapping platform, entering early access this year, can interrogate billions of individual protein molecules using fluorescently labeled probes, identifying proteins and proteoforms at the single-molecule level through machine learning. This kind of resolution was not possible even a few years ago.

Second, computational peptide design is accelerating. Tools like LigandForge can now generate over 700 candidate peptide sequences per second per GPU and produce predicted sub-100 nM binders across the majority of tested receptor targets.

Third, AI can learn patterns across proteomic datasets that no human could identify.

But these tools have real limitations. Peptide therapeutics face delivery challenges, including rapid degradation, limited membrane permeability and difficulty reaching intracellular targets. They exhibit higher specificity and fewer off-target effects than small molecules, but "fewer" is not "zero." And proteomic profiling, while advancing rapidly, is still far from continuous real-time monitoring of an individual's full proteome.

The Missing Piece Is Data

All of these tools are only as good as the data they run on. What we need is longitudinal clinical data tied to real patient outcomes, collected continuously, across diverse populations. That dataset doesn't exist yet at scale.

The concept of a "digital twin" of a patient's biology is gaining traction in research, with early work in areas like sepsis showing that proteomic data combined with AI can predict patient outcomes and identify high-risk individuals. But building medically relevant digital twins at a population scale requires the kind of structured, longitudinal data that current healthcare infrastructure doesn't generate.

Where The Data Actually Comes From

The data has to come from somewhere. The most natural source is the place where most clinical decisions already get made: primary care. It's the one point of continuous contact in healthcare, where relationships accumulate over time, and every interaction leaves a record.

That's the position Lotus AI is building from. Every patient interaction generates structured clinical data. Every outcome is a label. Over time, that compounds into the dataset the tools described above are waiting for, not collected in a lab or a clinical trial, but through the ordinary rhythm of ongoing patient care.

The companies that define this next era may not be the ones with the best protein-folding algorithms. They may be the ones who built trusted, continuous patient relationships and had the infrastructure to learn from them.

That's the bet we're making. And the proteomics revolution will prove it right.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?