China's pipeline does not support a simple winner-takes-all contest. The more defensible conclusion is segmentation: patient-specific vaccines, shared-mutation vaccines, tumor-associated-antigen products and viral-antigen products solve different biological and operational problems.

Among the Chinese programs reviewed, Everest Medicines is advancing the patient-specific EVM16 alongside the fixed-antigen EVM14, while Abogen Bioscience is advancing the personalized ABO2109 alongside shared/recurrent-mutation programs such as ABO2102 and ABO2013. Shanghai Xinpu's XP-004 regimen goes further by sequencing a fixed KRAS vaccine before a patient-specific multi-neoantigen vaccine.^{9–16, 20–21}

Two product architectures — but four antigen strategies

A true personalized cancer vaccine starts with the patient. Tumor and matched-normal samples are sequenced; tumor-specific mutations are identified; candidate neoantigens are ranked for expression, clonality, HLA presentation and predicted immunogenicity; and a unique vaccine is manufactured for that patient. The architecture broadens target matching to the individual tumor, but it retains prediction, tissue-quality, presentation, manufacturing and time-to-dose risk.^1–5

An off-the-shelf product is manufactured in advance without patient-specific redesign. Eligibility may depend on a recurrent mutation, viral status, antigen expression or, in some trials, tumor type without a mandatory patient-level antigen test. Its three biological strategies are shared/recurrent tumor mutations, tumor-associated antigens and viral antigens. It shifts personalization from manufacturing to patient selection; it does not mean "for everyone."

Sources: program descriptions, trial records and official disclosures.^6–23

Tumor mutational burden helps — but does not decide

A high tumor mutational burden can enlarge the pool of candidate neoantigens, but mutation count alone is an incomplete gatekeeper. A useful vaccine target must also be expressed, sufficiently clonal, processed, presented by the patient's HLA molecules and capable of producing a functional T-cell response.

The clearest China-linked peer-reviewed evidence comes from iNeo-Vac-P01, a personalized peptide — not mRNA — vaccine. A 2020 pan-cancer study reported feasibility, tolerability and vaccine-induced T-cell responses. A 2021 retrospective analysis (NCT03645148) evaluated iNeo-Vac-P01 with GM-CSF in seven patients with low-TMB advanced pancreatic cancer; mean PFS was 3.1 months and mean vaccine-associated OS was 8.3 months — not medians. The small, uncontrolled cohort cannot establish efficacy.^2–3

Rojas and colleagues' 2023 Nature paper examined a different product: autogene cevumeran, an individualized mRNA–lipoplex vaccine; its 2025 follow-up described long-lived vaccine-induced CD8+ T cells. The non-randomized study combined surgery, atezolizumab, vaccination and mFOLFIRINOX, supporting immunogenicity — not definitive vaccine-attributable benefit. It should not be conflated with the 2021 iNeo-Vac-P01 peptide-vaccine study.^4–5

The disease clock may matter more than the platform

Personalized manufacturing creates a period between tissue acquisition and first dose. The U.S. National Cancer Institute has described a typical personalized mRNA-vaccine production period of approximately one to two months, although actual timelines vary by platform, site and release-testing workflow.¹

That interval may be workable after complete resection, when the goal is to eliminate minimal residual disease and prevent recurrence. It is harder to tolerate in rapidly progressing, treatment-refractory disease. Off-the-shelf inventory offers an operational advantage in that setting — but only when a validated shared target exists and the patient's tumor carries it.

Illustrative personalized mRNA-PCV platform timeline

The workflow below adapts a China PCV manufacturing resource into an investor-oriented timeline. It is an indicative operating model rather than a validated industry benchmark: several activities may overlap, and actual turnaround depends on tissue quality, sequencing logistics, algorithm readiness, manufacturing capacity, release specifications and rework. The resulting six-to-ten-week range is directionally consistent with the NCI's approximately one-to-two-month description.¹

For diligence, the more useful metric is not the quoted best-case turnaround. Buyers should request the median and range from tissue receipt to release, the percentage of consented patients who reach dosing, failure and rework rates at each stage, and the proportion of batches delivered inside the intended clinical window.

• Patient-specific vaccines trade broader tumor matching for prediction, chain-of-identity, individualized CMC and time-to-dose risk.
• Shared-antigen vaccines trade rapid deployment and repeatable manufacturing for biomarker-coverage, expression, presentation, tolerance and escape risk.
• Cost follows the same structure: personalized products carry per-patient sequencing, bioinformatics, manufacturing and release costs; standardized products can spread batch costs across a target-positive population.

Why one company would build both

Everest: shared infrastructure, different products

Everest's EVM16 is a patient-specific mRNA–LNP program. In an official AACR 2026 disclosure, the company reported neoantigen-specific T-cell responses in eight of nine treated patients, one confirmed partial response and two cases of stable disease. All patients received two EVM16 monotherapy doses before EVM16 was combined with tislelizumab, so the clinical activity cannot be attributed to the vaccine alone. The evidence is early, conference-stage and based on nine patients.^9–10

EVM14 takes the standardized route: a fixed vaccine encoding five tumor-associated antigens for selected squamous cancers. Official disclosures and trial registration support U.S. and China IND clearance and global Phase I/IIa development. They do not yet establish efficacy.^11–12

Abogen: personalized discovery plus shared/recurrent mutations

Abogen's registered portfolio shows a similar segmentation. ABO2109 is patient-specific, whereas ABO2102 is a standardized KRAS neoantigen vaccine and ABO2013 targets recurrent EGFR mutations. ABO2102 now has both a hospital-sponsored IIT record (NCT06577532) and a company-sponsored Phase I record (NCT07455617). ABO2013's public evidence includes first-in-human Phase I findings from ChiCTR2300071001 disclosed in an AACR 2025 abstract. These remain preliminary conference findings from an early dose-escalation study, not peer-reviewed or controlled evidence of efficacy.^{13–16, 26–27}

The strategic logic is platform leverage rather than indecision. Antigen selection, mRNA engineering, lipid nanoparticle formulation, analytics and immune monitoring can support both architectures. Personalized programs test the full design-and-manufacture system; fixed-antigen programs convert selected targets into repeatable products with a more conventional CMC and inventory model.

XP-004 from Shanghai Xinpu BioTechnology tests a hybrid model

XP-004 is the clearest challenge to the competitor narrative. In a Phase I investigator-initiated study after radical resection of pancreatic cancer in chemotherapy-intolerant patients, the regimen began with four cycles of a fixed KRAS-targeted mRNA vaccine and then moved to nine cycles of a patient-specific multi-neoantigen vaccine, together with toripalimab.^20–21

At the 15 December 2025 cutoff, the conference abstract reported 16 treated patients, all recurrence-free after a median postoperative follow-up of 43.9 weeks; all 13 patients evaluable for immunogenicity developed neoantigen-specific T-cell responses. This is a notable early signal, not proof of efficacy: the study is small, single-arm, post-resection, short in follow-up and combined with PD-1 blockade. It cannot determine the contribution of the fixed KRAS component, the personalized component, toripalimab or patient selection.^20–21

What this means for financing and licensing

An off-the-shelf program can fit a conventional asset thesis: defined formulation, defined patient-selection strategy, repeatable manufacturing and a clearer path to inventory. Its commercial risk concentrates in target prevalence, antigen expression, HLA presentation, competitive intensity and the durability of target-positive patient selection.

A personalized program is closer to licensing a coordinated operating system than a single molecule. The investable asset includes the algorithm, data, tissue workflow, manufacturing network, quality system, clinical execution and the ability to reproduce all of them across sites and countries.

Coexistence is the more likely outcome

China's pipeline does not yet show that either architecture consistently improves outcomes. Most programs remain early, and the evidence base is dominated by small trials, registries, conference abstracts and company disclosures.

It does show that the architectures are being developed around different constraints. Patient-specific vaccines may be favored when adequate tissue is available, no sufficiently broad shared target exists, multiple credible neoantigens can be identified and the treatment window can accommodate individualized manufacture. Standardized vaccines may be favored when a shared or recurrent tumor mutation, tumor-associated antigen or viral antigen defines a meaningful population and rapid deployment or manufacturing scale matters.

Everest and Abogen are building both. XP-004 is testing both within one treatment sequence. The most plausible future is therefore coexistence by indication, antigen biology and disease timing — with selective hybrid regimens — rather than one platform displacing the other.

The real competition is not personalized versus off the shelf. It is between programs that align antigen strategy with the patient's tumor and clinical clock, and programs that do not.