A running collection of questions I find genuinely interesting — usually worked through with my grandfather over a series of long, winding conversations, and a good deal of AI brainstorming along the way. First entry: what mirror-image biology might mean for the field I actually work in.
Nearly every protein in every living thing on Earth is built from L-amino acids. It's a near-universal rule, and it's why a mirror-image version of a natural protein — built entirely from D-amino acids instead — looks chemically identical to the body but is geometrically backwards. Most natural enzymes, including the proteases that normally break proteins down, can't recognize or cut a backwards molecule. It just doesn't fit.
That single fact is the seed of an entire field: if a peptide can be built as its own mirror image, it can potentially resist degradation almost entirely — circulating longer, surviving harsher environments, and doing its job without being chewed up by the body's usual defenses.
The question my grandfather and I kept circling back to: if that's true for proteases, is it also true for the immune system's other major recognition system — the one T-cells use to "see" what's happening inside a cell in the first place?
Researchers are using D-amino acid versions of disease targets as "bait" to discover D-peptide drug candidates — a technique called mirror-image phage display, now accelerating thanks to better chemical protein synthesis.
A macrocyclic peptide built mostly from D-amino acids was developed against MMP7, a protein implicated in multiple cancers — and it held up far better against enzymatic breakdown than a standard peptide would.
A D-peptide-based imaging and therapy agent was shown to improve tumor targeting and antitumor immune activity when combined with checkpoint blockade — D-peptide chemistry showing up directly in immunotherapy research.
Olaptesed pegol (NOX-A12), a mirror-image RNA aptamer, is in active clinical trials — including a Phase 1/2 glioblastoma trial pairing it with radiotherapy, and completed Phase IIa studies in leukemia, blocking a chemokine that helps tumors exclude T-cells.
New computational models can take a target protein and generate candidate D-peptide binders directly — folding AI-assisted molecular design into a field that, until recently, depended entirely on slow, manual chemistry.
This is the part that's specific to what I do. T-cell receptor recognition of antigens presented on MHC molecules is famously sensitive to structure — even a single L-to-D amino acid swap in a known epitope has been shown to disrupt TCR recognition in published immunology research. That makes me wonder about something more deliberate.
If you took a patient-derived tumor organoid — completely standard, completely natural — and loaded it with the D-amino acid version of a known tumor antigen instead of the native one, would co-cultured T-cells still activate at all? Or would the antigen-processing and MHC-loading machinery simply fail to engage it?
If the T-cells never respond, that's not just a curiosity — it's a clean, structural way to isolate exactly which step of antigen presentation actually requires native chirality, using a completely ordinary organoid-immune co-culture system already common in the field.
It's worth being precise about scope. Everything above concerns individual D-peptides and proteins — non-living molecules used as drugs or research tools. That work is active, encouraged, and already in clinical trials.
That's a completely different question from building a whole self-replicating mirror-image organism — a "mirror cell" with its own mirrored genome and metabolism. In late 2024, a group of 38 senior scientists published a paper in Science warning that such organisms, if ever created, could evade immune defenses across humans, animals, and plants simultaneously, with potentially irreversible consequences — and called for that specific line of research to be halted.
So this notebook entry stays squarely in the first category. Mirror-image biochemistry, used at the molecule level, is one of the more genuinely promising directions in cancer drug design. Building mirror life is a different, much darker question — and one the scientific community has already, rightly, drawn a hard line against.
Entry 01 asked: what if we could systematically map how much of immune recognition depends on molecular chirality? That's a good question. But the real insight isn't just answering it — it's recognizing that nobody else has framed the question this way yet. The opportunity isn't to run a single three-phase experiment and publish a paper. It's to establish a new *subfield* and claim early intellectual ownership of how it's defined.
Full paper: For the complete technical framework including all three improvements (stereochemical tolerance thresholds, concrete AI workflows, Chiral Scan design), see the full Mirror Immunology research paper.
Instead of "let's test whether D-peptides work in organoid co-cultures," the framing is: let's build a comprehensive atlas documenting how chirality affects each stage of immune function. Which immune functions are chirality-dependent, and which aren't? That question is big enough to outlast any single experiment or lab. It's the kind of question that other researchers can contribute to, cite, and build on. And if Caroline publishes the framework first, she's defined the terms.
The key moves here:
1. The distinctive contribution isn't D-peptides themselves. Many groups are already studying D-amino acids and mirror-image therapeutics. The novel part is using patient-derived tumor organoids plus AI-designed mirror molecules to *systematically map* the exact chirality requirements of antigen processing, MHC loading, presentation, and T-cell recognition. That framing elevates this from "we tested something" to "we're building a resource the field needs."
2. Position-dependent tolerance matters. Anchor residues (directly contacting the MHC groove) are fragile to D-substitution; solvent-exposed residues can tolerate it while abolishing TCR recognition. That asymmetry is the conceptual hook — it means chirality-dependence isn't binary, it's position-specific and mappable. Any future work in this space will need to account for this distinction.
3. Add "Chiral Scan" peptides as a third experimental condition. Beyond just WT (L-peptide) vs. fully inverted (all-D peptide), introduce peptides where only a single critical amino acid is flipped. This isolates the exact atomic contact points where recognition breaks down and makes the experimental architecture far more robust.
4. Leverage AI from day one. Rather than synthesizing candidates and testing them serially, use computational tools — AlphaFold 3, D-Flow diffusion models for non-canonical amino acids, molecular dynamics simulations — to predict which positions will tolerate D-substitution. That moves from discovery-by-experiment to discovery-informed-by-prediction. It's faster, cheaper, and generates the data that will populate the Atlas.
Both technical reviewers independently flagged the same three gaps in the draft paper. These aren't optional refinements — they're the difference between "interesting idea" and "publishable framework." Before the website essay or preprint goes public, make sure these three are locked in:
What to add: Explicitly state that chirality is not an all-or-nothing switch. Anchor residues (directly contacting the MHC groove) are fragile to D-substitution and often lose binding entirely. Solvent-exposed residues can tolerate D-substitution without destroying MHC binding, yet completely abolish TCR recognition. This position-specific asymmetry is the conceptual hook that makes the whole framework work.
What to add: Replace generic "AI systems" language with specific tools and frameworks. Name: AlphaFold 3 for structure prediction, D-Flow diffusion models for designing non-canonical amino acid variants, molecular dynamics simulations for modeling peptide-MHC-TCR complexes. This moves from "AI could help" to "here's exactly which computational tools you'd use and why."
What to add: Beyond L-peptides (control) and all-D-peptides (experimental), introduce a third intermediate condition: "Chiral Scan" peptides where only a single, critical amino acid is flipped to D-form. This isolates exact atomic contact points where recognition breaks down and makes the experimental architecture far more robust and interpretable.
Publish a polished essay on your site (or Medium, Substack — wherever reaches researchers) laying out the Mirror Immunology concept, why it matters, and what a systematic atlas would look like. This is fast, builds visibility, and creates a citable artifact that demonstrates you thought about this first. Faster than peer review, creates immediate proof-of-concept visibility.
Post a more rigorous version to arXiv or bioRxiv as a formal perspective/hypothesis piece. Include the technical nuances — stereochemical tolerance thresholds, the specific computational tools (AlphaFold 3, D-Flow), the "Chiral Scan" experimental design. This gives the idea permanence and establishes a timestamp for intellectual ownership. Researchers working on related projects will cite it.
Target an invitation or submit a short commentary to a high-profile immunology or synthetic biology journal — Nature Immunology, Immunity, Trends in Immunology, or Synthetic Biology. By this point, the idea has circulated and been discussed. A journal perspective in a recognized venue positions Caroline as an early voice who helped define the field before it had a name.
Before diving into heavy bench work, connect with a computational biologist who specializes in structural modeling and protein design. Have them simulate what happens when D-amino acids sit inside real MHC molecules and how TCRs would bind (or fail to bind) the distorted complex. This generates preliminary computational data that informs which experiments are most likely to yield insight — and potentially attracts a collaborator who might co-lead the first pilot study.
Why this sequencing works
The goal isn't to hide the idea until you've published it. The goal is to stake your claim in the intellectual landscape *before* the field recognizes it as a field. By the time other labs start designing D-peptide + organoid experiments, Mirror Immunology as a named, framed research direction will already be associated with your thinking. You're not racing to be first with an experiment; you're racing to be first with a *framework* that everyone else will reference.
The beauty of this path is it doesn't require you to commit lab resources upfront. A website essay takes a weekend. A preprint takes a few weeks of writing and gets feedback before any experiments run. A journal perspective is short and positions you as a thought leader, not just a technician. Only after the framework is established and you've found a computational partner do you move to Phase 5: the actual pilot study.
This isn't about asking the lab for resources today. It's about establishing yourself as an independent thinker who can identify gaps in how the field asks questions. Publishing an essay or perspective on your own time, building a collaboration with a computational biologist, and generating preliminary computational data — none of that requires the lab's bench, budget, or approval. It's intellectual legwork that demonstrates research maturity.
When you're ready for the actual pilot study, you'll approach Dr. Chan with: a published framework other researchers are already discussing, preliminary computational data suggesting which experiments will be most informative, and a computational collaborator ready to partner. That's a much stronger position than walking in with a notebook idea and a three-phase experiment plan. It signals you can identify a scientific opportunity, build the conceptual case, and assemble collaborators — exactly what makes a good independent researcher.