S13. State of the science by technology family

This appendix sets out the state of the science in September 2026 for each technology family that could supply alternative protein from Vietnam. For each family it gives the key numbers with their DOIs, what the evidence means for Vietnam, and where our earlier drafts were wrong. Technology fit draws the conclusions; Costs and benchmarks holds the full cost inputs; S7. Research landscape holds the Vietnamese research record.

Most of the evidence here is global. Vietnamese data exist only for a few families (aquafeed trials, side-stream protein extraction, microalgae cultivation). Where evidence is absent, we say so.

#S13.1 Overview: where each family stands in 2026

The "Vietnam evidence base" column summarises the Vietnamese peer-reviewed record found in wave 1 (Scite) and wave 2 (OpenAIRE), detailed in each section below. Counts are lower bounds ^1,2 VN-directTrung bình.

#S13.2 Plant proteins and texturisation

#S13.2.1 State of the science in 2026

• Protein quality of common isolates and concentrates is covered in section S13.3. Soy isolate is the reference: measured pig DIAAS of 97 on the older child and adult pattern ³ generalCao.
• Costs. We found no verified capex or opex benchmark for plant protein isolation or high-moisture extrusion. The only Southeast Asian TEA found (a Thai study of freeze-aligned meat analogue, doi:10.1016/j.fufo.2023.100269) had no readable text ¹⁶ VN-adjacentThấp.
• Indicative Vietnamese cost of textured soy protein (TVP). Our wave 1 worked example puts dry TVP from soy flour or concentrate at USD 790 to 2,070 per tonne ex-factory (about 3,700 to 10,700 VND per 100 g of protein). Imported soy flour or concentrate is about 70% of the cost. This is our calculation from sourced Vietnamese input prices and assumed equipment costs ^17,18,19 VN-directThấp. See Costs and benchmarks.

#S13.2.2 Vietnamese evidence

• Side-stream plant protein is Vietnam's most developed line. Rice-based dried distillers' grain (DDG) from Vietnamese ethanol plants holds 55 to 80% protein on a dry basis; cassava DDG holds only 13 to 16% (2024 preprint, doi:10.21203/rs.3.rs-3171967/v1) ²⁰ VN-directTrung bình. Alkaline extraction recovered 90% of rice DDG protein, and enzymatically extracted protein was 77.1% digestible in vitro (doi:10.1002/apj.70043) ²¹ VN-directCao.
• Other extraction work: ultrafiltration recovery of rice protein (doi:10.1016/j.ifset.2021.102692) and rice bran protein isolate (2016) ^22,23 VN-directTrung bình.
• Products. Replacing 10 to 15% of meat with germinated mung bean flour was accepted in sausage (doi:10.1111/ijfs.17089); mangosteen peel extract extended the shelf life of soy burgers (doi:10.17113/ftb.63.01.25.8629) ^24,25 VN-directCao. A 2026 review of plant-based fish analogues has a Vietnamese first author whose affiliation was not established ²⁶ VN-directThấp.
• Counts. The OpenAIRE query for plant-based meat returned 9 Vietnam-tagged records for 2015 to September 2026; we first judged about 6 to be on topic. One of them, a 2025 study of the texture of meat-analogue extrudates (doi:10.1093/ijfood/vvae080), turned out to come from a Taiwanese university, leaving about 5 ^2,4 VN-directTrung bình.
• Equipment. We found no extruder for food texturisation in the Vietnamese literature or on the pages of pilot facilities we read ^1,27 VN-directTrung bình. No local maker of extruders was confirmed, and the Vietnam presence of the main foreign vendors could not be verified ^28,29 VN-directTrung bình.

#S13.2.3 What it means for Vietnam

• Production cost is not what blocks textured plant protein in Vietnam. On our indicative estimate, the ex-factory cost of TVP protein is roughly 4 to 19 times below the retail price of meat and egg protein, so product quality, demand and distribution margins decide ^30,31,32 VN-directThấp.
• Vietnam's cost edge is small because the main input, soy, is imported: Vietnam imported 2.61 Mt of soybeans in 2025, 98.4% of supply ^33,34 VN-directCao.
• The knowledge gap is specialist: extrusion, texturisation and sensory structure work are almost absent from the Vietnamese literature ^1,2 VN-directTrung bình. Rice DDG and broken rice are the obvious local protein bases to texturise.

#S13.3 Protein quality and DIAAS

#S13.3.1 State of the science in 2026

DIAAS (Digestible Indispensable Amino Acid Score) is the current FAO method for protein quality. The older-child pattern (over 3 years, adolescents and adults) applies to school-age children; the young-child pattern (6 to 36 months) applies to complementary foods ³⁵ generalCao. Values below come from ileal digestibility measured in growing pigs unless stated. The table also covers the foods Vietnamese meals and plant products actually contain (the full table, with methods and notes, is protein_quality.csv).

• Processing and method move the numbers. Soy products range from DIAAS 64 (tofu) to 91 (soy milk) in the one quantitative review, and textured soy from 65 to 97; PDCAAS-like scores flatter plant proteins against DIAAS ^38,44 generalTrung bình. In vitro methods give tofu 98 ³⁹ generalTrung bình. The two tofu figures use different patterns and methods, so we give both and do not average them (DG-320 in R2. Disagreements).

• Blends keep quality high. DIAAS values of single foods add up in mixed meals ^35,43 generalCao, so a protein-weighted average of component scores is a floor for a blend. A minced-pork dish with 20 to 30% of its protein from textured soy scores at least 97 on the older-child pattern and at least 90 on the young-child pattern, with zinc per gram of protein 12 to 19% below all-pork (our calculation) ^38,43,45,46 generalTrung bình. A plant dish that removes all meat is a different case: it needs added zinc (below).

• Zinc and phytate matter as much as amino acids. Soybean phytate gives a phytate-to-zinc molar ratio of about 34 (9 to 43), above the 15 at which Vietnam's 2025 school-meal guidance rates zinc absorption as poor, and most soy-milk phytate is taken into the tofu curd ^47,48,49,45 generalTrung bình. Soy iron is also largely unavailable without phytate reduction ^50,51 generalTrung bình. Plant foods carry no vitamin B12.

• Human data differ from pig data. In Indian children under 2 years, true ileal digestibility of indispensable amino acids (dual isotope method) was 65.2% for mung bean, 78.5% for rice and 87.4% for egg. Mung bean digestibility was about 20% lower than pig estimates (doi:10.1093/ajcn/nqy265) ⁵ VN-adjacentCao. Typical cereal-legume complementary diets had a DIAAS of about 80; adding egg or milk raised it to 100 in calculation ⁵ VN-adjacentCao.

• Rice and legume blends. We found no measured (in vivo) rice plus legume blend with a DIAAS of 100 or more. Claims of 100 come from calculations, for example pea plus rice blends reaching a PDCAAS of 1.00 on the adult pattern (doi:10.1139/apnm-2021-0806) ⁵² generalTrung bình.

• Microalgae and duckweed. No DIAAS for microalgal biomass was available as of a 2021 review (doi:10.3390/foods10123002) ¹⁰ generalTrung bình. For duckweed we found only amino-acid scores and in vitro digestibility (section S13.8), and no PDCAAS or DIAAS study ^53,54 generalThấp.

#S13.3.2 Vietnamese evidence

• We found no Vietnamese DIAAS or PDCAAS study ¹ VN-directTrung bình, and no DIAAS measured in school-age children anywhere; a single OpenAlex search found no ileal digestibility study of soy or tofu in that age group (Low-confidence negative). In vitro digestibility has been reported, for example 77.1% for rice DDG protein ²¹ VN-directCao.
• No Vietnamese laboratory page we read lists an amino-acid profile or a protein-digestibility assay among its services; PDCAAS and DIAAS work will probably need an overseas laboratory for now ^55,56 VN-directTrung bình.

#S13.3.3 What it means for Vietnam

• Mung bean is a strong local legume by pig DIAAS but weaker in humans; rice protein alone is poor because it is lysine-limited ^3,41,5 generalCao. Prefer human data for food claims (DG-110).
• A Vietnamese DIAAS reference dataset (rice DDG, mung bean, spent yeast, catfish side-stream isolates, spirulina) is a cheap public good that any novel-food dossier will need. The list includes tofu, textured soy and blended minced dishes, measured on both patterns, with phytate, zinc and iron; this is why policy option 13 (PO-027) now includes phytate-to-mineral analysis in the laboratory scope (Policy options).
• For school-age children the binding nutrients are zinc and iron, not protein quality; protein quality matters most for plant-only meals for older boys and, much more, for children under two, where plant products face the young-child pattern and registration as foods for children up to 36 months (Decree 15/2018) ^45,57,58 VN-directTrung bình. A 20 to 30% soy extender in minced dishes raises no protein-quality objection. A zinc-fortified textured-soy school dish is a research option, not a recommended product: in rural Vietnam, biscuits fortified with zinc and iron cut anaemia and zinc deficiency in children aged 6 to 8 ⁵⁹ VN-directCao, but no trial has tested a fortified soy dish.

#S13.4 Fungal biomass, mycoprotein and koji

#S13.4.1 State of the science in 2026

• RNA reduction is a fixed cost. Mycoprotein RNA falls from about 10% to about 2% of dry weight after a 68 C, 20-minute heat step that lets the fungus's own enzymes break down RNA. The step is described as industrially expensive (doi:10.3390/nu12041115) ⁶⁰ generalCao. A secondary review says below 1% is needed "to be considered safe"; we use the primary figure (DG-107) ⁶¹ generalTrung bình.
• Protein content. Fusarium venenatum mycoprotein is about 30 to 45% protein on a dry basis (secondary review, doi:10.3390/fermentation11010024) ⁶¹ generalTrung bình.
• Aspergillus oryzae safety. A. oryzae cannot make aflatoxin, because its aflatoxin gene cluster carries disabling mutations. It can make cyclopiazonic acid (CPA) and 3-nitropropionic acid. Modern industrial strains carry a deletion that removes both the aflatoxin and CPA clusters. Kojic acid is not classed as a mycotoxin. Reports of aflatoxin-producing "A. oryzae" are misidentified A. flavus (doi:10.1007/s00253-018-9354-1) ⁶² generalCao.
• Fungal biomass on side streams. In a shake-flask study on diluted molasses vinasse, crude protein (N x 6.25) was 44.7% for A. oryzae, 57.6% for Neurospora intermedia and 50.9% for Rhizopus oryzae; F. venenatum and Monascus purpureus did not grow (doi:10.3390/fermentation5040099) ⁶³ generalCao. The biomass yield the paper reports cannot be read as a broth concentration (see the correction below).
• Costs. A full TEA of continuous airlift mycoprotein production (2 x 155 m3 reactors, 2,000 kg/h wet output) gives USD 3.55/kg wet, or USD 29.56/kg of protein, with capex of about USD 108 million; the reactors are 66.7% of capex (doi:10.3389/fsufs.2023.1204307) ⁶ generalCao. The authors conclude that mycoprotein can match beef on protein cost but "will not be an economic alternative for inexpensive products such as chicken". Buying food-grade biotin and zinc sulphate instead of laboratory grade cut cost by 22% ⁶ generalCao.
• Approvals. Singapore has approved five fungal biomass or mycoprotein ingredients: Nature's Fynd (2021), The Protein Brewery (2024), The Better Meat Co (2024), Infinite Roots (2025) and Ultimeat (2026) ¹² VN-adjacentCao. China reportedly issued its first mycoprotein approval in 2025 (Fushine Biotech); the date was not verified ^64,65 generalThấp.

#S13.4.2 Vietnamese evidence

• VNU Hanoi groups have built fungal genetic tools: a dual-auxotrophic transformation system in A. oryzae (doi:10.1007/s11274-021-03060-z) and 100% pyrG deletion efficiency in five Cordyceps militaris strains, the base for a food-grade expression platform (doi:10.47371/mycosci.2024.10.003) ^66,67 VN-directTrung bình.
• A 2025 study in the National Institute for Food Control journal tested aflatoxin in 38 samples of traditional fermented soy (tương) ⁶⁸ VN-directTrung bình.
• A 2025 preprint reports that solid-state fermentation of cassava and soybean residue with oyster-mushroom (Pleurotus ostreatus) mycelium raised protein 1.84-fold ⁶⁹ VN-directTrung bình.
• We found no Vietnamese study on mycoprotein or tempeh-type food fungi ^1,2 VN-directTrung bình.

#S13.4.3 What it means for Vietnam

• Fungal biomass on cassava or molasses residues is scientifically plausible, but published yields per litre must be read with care. Pilot data on Vietnamese substrates are needed before any cost model.
• A koji route that relies on the tương tradition still needs strain characterisation for CPA and 3-nitropropionic acid. Wild starter cultures are not automatically safe, and industrial deletion strains are proprietary ⁶² generalCao.
• RNA reduction and drying are unavoidable cost lines for food use; feed use tolerates higher RNA ⁶⁰ generalCao.
• Our indicative Vietnamese cost of fungal biomass protein for feed is USD 2,227 to 6,612 per tonne of product, or USD 4,050 to 14,700 per tonne of protein; capital charges and maintenance are 54 to 67% of cost (our calculation, wave 1 worked example) ^6,70,18 VN-directThấp. Fishmeal protein cost about USD 2,625 per tonne in 2025 and USD 3,846 in August 2026 ⁷¹ generalCao.

#S13.5 Yeast and bacterial single-cell protein

#S13.5.1 State of the science in 2026

Yeast.

• Candida utilis grown on cassava peel hydrolysate reached 47.5% crude protein on glucose, 49.1% on acid hydrolysate and 56.7% on enzymatic hydrolysate, with a biomass yield of 0.44 to 0.52 g per g of sugar and a maximum growth temperature of 41 C (doi:10.17265/2159-5828/2012.08.004) ⁷² VN-adjacentTrung bình. These are crude protein values (N x 6.25), which include nucleic-acid nitrogen, from a low-tier journal on Nigerian cassava peel.
• Brewer's yeast replaced 45% of fishmeal in hybrid "Thai Panga" catfish with better growth and immunity, according to a 2024 review (doi:10.3390/ani14192851) ⁷³ VN-adjacentTrung bình.
• China's Angel Yeast makes more than 11,000 t/yr of yeast protein at Yichang ⁶⁵ generalTrung bình.

Bacterial SCP in aquafeed. Single-cell protein trials are summarised below (details in replacement_trials.csv).

• Bacterial SCP can reach up to about 65% protein on a dry basis (secondary review) ⁶¹ generalTrung bình.
• A single trial is not the last word on methanotroph meal: FeedKind-specific shrimp trials were not retrieved ⁷ generalTrung bình.

Costs and scale.

• SCP from steel-mill off-gas via acetate: USD 4.15/kg (USD 2.78/kg if intensified), with USD 320 million of capital for 20,000 t/yr (doi:10.3390/fermentation9080771) ⁷⁰ generalCao.
• SCP from hydrogen and CO2: minimum selling price USD 2,070/t (abstract only, doi:10.1021/acs.est.3c10312) ⁸² generalTrung bình.
• SCP made with renewable electricity: EUR 5.5 to 6.1/kg of protein in 2028, EUR 4.0 to 4.5 in 2030 and EUR 2.1 to 2.3 in 2050 at optimal sites (doi:10.1038/s41467-025-56364-1) ⁸³ generalCao.
• Scale references: Solar Foods' Factory 01 can make at most 160 t/yr of Solein ⁸⁴ generalTrung bình; Calysseo runs a 20,000 t/yr FeedKind plant in Chongqing with two 10,000 t/yr fermenters ⁸⁵ generalTrung bình; ICI's 1,500 m3 Pruteen feed-SCP reactor ran from 1979 to 1987 and sold at about twice the price of soy protein ⁶ generalTrung bình.
• A Chinese bacterial biomass from glutamic-acid fermentation, at least 70% crude protein, is already imported into Vietnam for pig, poultry, fish and shrimp feed ⁸⁶ VN-directTrung bình.

#S13.5.2 Vietnamese evidence

• HUST studied nucleic-acid reduction in spent brewer's yeast hydrolysate (2018) and recovered 84.9% of the protein from spent brewer's yeast for a food product (2024) ^87,88 VN-directTrung bình.
• Spent brewer's yeast replaced up to 60% of fishmeal protein in giant freshwater prawn without loss of growth or survival (doi:10.1111/anu.12915) ⁸⁹ VN-directCao. Brewer's yeast has also been tested against soybean meal in Nile tilapia ⁹⁰ VN-directCao.
• A 2016 paper isolated a methane-oxidising bacterium for SCP, with no follow-up found ⁹¹ VN-directTrung bình. A 2023 conference paper on SCP from bamboo cellulose was retracted and is not used ⁹² VN-directCao.
• We found no Vietnamese trial of bacterial SCP, methanotroph meal or microalgae as bulk protein in pangasius or shrimp ¹ VN-directTrung bình.

#S13.5.3 What it means for Vietnam

• Shrimp feed is the right target for bacterial and yeast SCP, but the evidence supports partial replacement (about 30% of fishmeal per ingredient), not full replacement ^7,74 generalCao.
• Pangasius and tilapia feeds already use little fishmeal, so SCP there competes with soybean meal on price, a harder benchmark: soybean meal protein cost about USD 880 per tonne in August 2026 against USD 3,846 for fishmeal protein ⁷¹ generalCao.
• Early life stages need their own trials; the barramundi fry result is a warning ⁸⁰ generalTrung bình.
• SCP from tofu and soy-milk wastewater is a Vietnam-relevant lead worth a local trial, but it rests on one preprint ⁸¹ VN-adjacentTrung bình.
• For legal routes (livestock list, aquafeed trials, penalties since August 2026), see S9. Regulation.

#S13.6 Precision fermentation

#S13.6.1 State of the science in 2026

• The titre benchmark. The leading review states: "The cost-efficiency requirements involve production titers above 50 g/L" (Nielsen, Meyer and Arnau, Annual Review of Food Science and Technology 15:173-187, online 2023, issue 2024; doi:10.1146/annurev-food-072023-034256) ⁸ generalTrung bình. The same group repeats "protein titers well above 50 g/L are needed" in 2025 and gives cost targets of USD 10/kg by 2025 and USD 1/kg by 2035 as "envisaged" (doi:10.1177/15509087251398318) ⁹³ generalCao. Neither paper shows the cost model behind the 50 g/L figure: it is an expert assertion, not a calculated threshold.
• Industrial hosts. Industrial A. oryzae strains are the most mature hosts. A 2025 study on bovine beta-lactoglobulin raised product amount by 79% and carbon yield by 73% through pH, temperature and feed optimisation, but kept the absolute titre confidential ⁹³ generalCao.
• Best published leghemoglobin titres:
  • 3.5 g/L, secreted, in Pichia pastoris (Shao et al. 2022, doi:10.1016/j.biortech.2022.127884; value read through a citing paper) ^94,95 generalTrung bình;
  • 7.27 g/L, intracellular, in Kluyveromyces marxianus (Tian et al. 2024, doi:10.3389/fbioe.2023.1329016) ⁹⁵ generalCao;
  • 10.1 g/L in 5 L bioreactors, also in K. marxianus (Chen et al. 2026, doi:10.1002/biot.70239) ⁹ generalCao. These are 5 to 15 times below the bulk-protein benchmark. Heme proteins are used at only 0.5 to 2% of a product, so the benchmark binds less for them ⁹ generalCao.
• Trichoderma reesei. VTT produced beta-lactoglobulin and ovalbumin in T. reesei (Aro et al., Food Research International 163:112131, dated 2023; doi:10.1016/j.foodres.2022.112131), but we could not read the titres ⁹⁶ generalThấp. VTT reports brazzein (a sweet protein) at 1.3 g/L in a 250 mL bioreactor, with an estimated cost of EUR 57 to 80/kg; it also states more than 80 g/L for one unnamed heterologous protein as unpublished data, which we treat as a claim, not evidence (doi:10.3389/fbioe.2025.1688495) ⁹⁷ generalThấp.
• OPENPichia. A genome-sequenced Komagataella phaffii type strain (NCYC 2543) with a HOC1 truncation that restores transformability. It is near-identical to the patented strain NRRL Y-11430, and no strain was consistently better at protein production. It comes under "liberal end-user distribution licences" that allow royalty-free commercial manufacture, with a toolkit free of third-party material transfer agreements (doi:10.1038/s41564-023-01574-w) ⁹⁸ generalCao.
• Regulatory science. EFSA issued an opinion in 2024 on soy leghemoglobin from genetically modified K. phaffii as a food additive (colour in meat analogues) (doi:10.2903/j.efsa.2024.8822) ⁹⁹ generalTrung bình. Singapore has approved three precision-fermented ingredients: Remilk beta-lactoglobulin (2023) and two human milk oligosaccharides from Chr. Hansen (2025) ¹² VN-adjacentCao.
• Costs and capacity. We could not verify any peer-reviewed TEA for precision-fermented food proteins. For scale, Liberation Bioindustries' plant in Richmond, Indiana has 600 m3 of fermentation and starts operating in 2026 ¹⁰⁰ generalTrung bình.

#S13.6.2 Vietnamese evidence

• Vietnam has moderate strength in recombinant expression, mostly of pharmaceutical and enzyme proteins: growth factors FGF-2, KGF/FGF7 and PDGF-BB in Pichia (VNU-HCM, 2020 to 2021), an alkaline phytase aimed at aquafeed (2025) and a bacteriocin applied to pork bologna (2024) ^{101,102,103,104,105} VN-directTrung bình.
• After screening, Vietnam has zero records on precision fermentation or recombinant food proteins (dairy, egg, heme) for 2015 to 2026 ^2,1 VN-directTrung bình.

#S13.6.3 What it means for Vietnam

• A Vietnamese precision-fermentation startup using public strains would start 5 to 10 times below the bulk-protein titre benchmark. High-value, low-dose ingredients (heme proteins, sweet proteins, enzymes) are the only near-term fit ^8,9 generalTrung bình.
• OPENPichia removes one freedom-to-operate barrier for Vietnamese groups already working with Pichia; product-specific patents (for example on leghemoglobin constructs) still apply ⁹⁸ generalCao.
• K. marxianus links precision fermentation to thermotolerance (section S13.13), which suits Vietnam's climate ^95,15 generalCao.
• Legally, a precision-fermented protein is probably a "product of a GMO" in Vietnam, and a plant using a GM strain needs MAE's closed-production certificate ¹⁰⁶ VN-directTrung bình. See S9. Regulation.

#S13.7 Microalgae

#S13.7.1 State of the science in 2026

• No DIAAS for microalgal biomass was available as of a 2021 review ¹⁰ generalTrung bình. We did not verify spirulina protein content or digestibility in this study.
• In a whiteleg shrimp digestibility screen at 30% of the test diet, Chlorella did not reduce growth, while methanotroph meal and insect meals did ⁷ generalCao.
• Singapore has approved two Chlamydomonas reinhardtii algae biomass ingredients from Triton Algae Innovations (2019 and 2022) ¹² VN-adjacentCao.

#S13.7.2 Vietnamese evidence

• Microalgae is Vietnam's largest topic in the alternative-protein bibliometric set: 87 Vietnam-tagged spirulina records for 2015 to 2026 (3.62 per 10,000 publications), similar to Malaysia and Indonesia and below Thailand. But the most-cited Vietnamese spirulina papers are about wastewater, biofuel and nanoparticles; only one surfaced paper is about extracting spirulina protein as an ingredient ² VN-directTrung bình.
• A salt-tolerant spirulina strain reached 48.73% protein (dry weight) in 42 ppt medium (doi:10.3390/su151511906) ¹⁰⁷ VN-directCao. A 2021 review covers Arthrospira production in Vietnam ¹⁰⁸ VN-directCao. Microalgae isolated from Vietnamese shrimp ponds have been assessed for food, feed and biodiesel ¹⁰⁹ VN-directTrung bình.

#S13.7.3 What it means for Vietnam

• Vietnam has cultivation skills and saline-water sites, but protein quality data and ingredient processing are missing.
• Legally, microalgae food can be self-declared or registered as a health supplement; microalgae are on the livestock feed list but not on the aquafeed list ^110,111,112 VN-directCao.

#S13.8 Duckweed and aquatic plants

#S13.8.1 State of the science in 2026

• Composition. Wolffia protein is 20 to 30% of freeze-dried weight across 11 laboratory-grown species (doi:10.3389/fchem.2018.00483) ⁵³ generalCao. Thai samples held 44.3% crude protein in Wolffia, 22.2% in Lemna and 18.0% in Spirodela; in vitro standardised ileal protein digestibility (pig-simulating) was 72% for Lemna, 69% for Wolffia and 39% for Spirodela (doi:10.48048/tis.2024.8324) ¹¹³ VN-adjacentTrung bình. Drum-dried Wolffia globosa held 32 to 34% protein, with in vitro digestibility rising from 81% fresh to 85% dried (doi:10.1002/jsfa.70332) ¹¹⁴ VN-adjacentTrung bình.
• Manganese is the documented food-safety issue. Wolffia samples held 78 to 431 mg of manganese per kg of dry weight ⁵³ generalCao. EFSA could not establish the safety of whole Wolffia globosa powder (2021, doi:10.2903/j.efsa.2021.6938) or of mixed water-lentil powder (2021, doi:10.2903/j.efsa.2021.6845) because of manganese intake. It found a Lemna protein concentrate safe under the proposed uses (2023, doi:10.2903/j.efsa.2023.7903) ^115,116,11 generalCao. Protein extraction removed the problem that blocked whole-plant powders.
• Human data. A human study found Lemna minor protein absorbed less well than pea protein (doi:10.1007/s11130-022-00952-9) ¹¹⁷ generalTrung bình.
• Productivity. About 30 t of dry matter per hectare per year is the average cited in a 2026 review, and 33 t per hectare was reached over 8 months in domestic wastewater; "up to 100 t per hectare per year" is an extrapolated upper bound ^118,119 generalThấp. We use 10 to 30 t per hectare per year for planning until Vietnamese field trials exist (our judgement; DG-109).

#S13.8.2 Vietnamese evidence

• A national survey collected more than 100 wild Spirodela, Lemna and Wolffia samples and keeps them in a laboratory collection (2022) ¹²⁰ VN-directTrung bình. A Vietnamese-led team assembled a chromosome-scale Spirodela intermedia genome (2020) ¹²¹ VN-directCao. Duckweed removed nutrients from anaerobically treated swine wastewater in laboratory ponds ¹²² VN-directCao.
• We found no protein extraction, food trial or feed trial with Vietnamese duckweed from 2015 to 2026 ¹ VN-directTrung bình. The OpenAIRE duckweed cluster (23 Vietnam-tagged records) is mostly genomics and taxonomy ² VN-directTrung bình. A European researcher's duckweed work linked to Vietnam in earlier leads has no Vietnamese publication ¹²³ VN-directTrung bình.

#S13.8.3 What it means for Vietnam

• Duckweed grown on aquaculture effluent fits feed. For food, the EU record shows that mineral control (manganese) and controlled water are the gate, and that a protein concentrate is an easier regulatory path than whole-plant powder ^115,11 generalCao.
• Legally, bèo (duckweed) is on the livestock feed list but not on the aquafeed list ^111,112 VN-directCao.
• Vietnam holds germplasm and a genome but has not used either for protein. Screening the national collection for protein and manganese is a quick public good.

#S13.9 Seaweed

• State of the science. The wave 2 science review did not cover seaweed protein, so we give no global benchmark here.
• Vietnamese evidence. Green seaweed (Cladophora) protein partly replaced fishmeal protein in black tiger shrimp postlarvae (doi:10.1007/s10811-018-1457-7), and gut weed (Ulva intestinalis) meal was evaluated with Artemia biomass as dietary protein for the same species (doi:10.1016/j.ejar.2022.11.003) ^126,127 VN-directCao. A 2022 Wageningen report studied seaweed cultivation in Vietnam for livestock methane reduction ¹²⁸ VN-directTrung bình. We found no Vietnamese work on food protein extraction from seaweed ¹ VN-directTrung bình; the OpenAIRE seaweed-protein query found one Vietnam-tagged record ² VN-directTrung bình.
• What it means for Vietnam. Seaweed (Ulva, Caulerpa, Kappaphycus) is a possible feed and food protein source with almost no Vietnamese protein data. Rong (seaweed) is on the livestock feed list ¹¹¹ VN-directCao. We treat seaweed protein as an open research question, not a near-term supply option.

#S13.10 Mushrooms and mycelium

• State of the science. Mycelium and fungal biomass foods have reached approval: Singapore lists Pleurotus pulmonarius mycelium biomass (Infinite Roots, 24 September 2025) and two Neurospora crassa mycoproteins (The Better Meat Co, 15 October 2024; Ultimeat, 27 April 2026) ¹² VN-adjacentCao. The wave 2 science review did not cover mycelium growth or texture data.
• Vietnamese evidence. Vietnam has about 40 papers on mushroom cultivation and bioactives but only 2 to 3 on protein ¹ VN-directTrung bình. Examples: grey oyster mushroom grown on cassava peel from starch production (doi:10.1051/e3sconf/202340501009); dual-enzyme Pleurotus pulmonarius protein hydrolysates (doi:10.1016/j.fochx.2025.103310); and the 1.84-fold protein gain in cassava and soybean residue fermented with Pleurotus mycelium (preprint) ^129,130,69 VN-directTrung bình. The Cordyceps toolkit (section S13.4) is a possible food-grade expression platform ⁶⁷ VN-directTrung bình.
• What it means for Vietnam. Edible-mushroom species carry lower legal risk than new strains under the current food law, because they have food history ^131,110 VN-directTrung bình. Mycelium grown on cassava or soybean residue links Vietnam's mushroom skills to its side streams, but food-grade protein data are thin.

#S13.11 Cultivated meat and seafood

#S13.11.1 State of the science in 2026

• Cell lines for Vietnamese species. Striped catfish (Pangasianodon hypophthalmus, Vietnamese tra) cell lines exist in India (ICAR-NBFGR): a thymus line, PHT (accession NRFC-078; 52 passages; L-15 medium with 20% fetal bovine serum; 28 C; epithelial and adherent) (doi:10.56093/ijans.v93i2.128796), a gill line (2022, doi:10.1007/s10695-022-01053-9) and a caudal fin line (2018, doi:10.1016/j.actatropica.2018.03.015). A skin epithelial line was published from Taiwan in 2025. All were built for virology and toxicology, not food ^{132,133,134,135} VN-adjacentCao.
• Shrimp. "PmLyO-Sf9" is a black tiger shrimp lymphoid cell x insect Sf9 hybrid from Cochin University of Science and Technology (CUSAT), used for virus work (doi:10.1016/j.fsi.2021.03.023). It contains insect genetic material. We found no true continuous shrimp cell line in the literature ^136,137 generalCao.
• Density limits (Humbird 2021, doi:10.1002/bit.27848). Absolute maximum 258 g/L of wet cells; 110 g/L in fed-batch before ammonia inhibition; 140 g/L with one perfusion filter; 195 g/L oxygen-limited. With inefficient metabolism, inhibition caps density at 7 to 20 g/L. The conclusion is that economics "would likely preclude" affordability as food ¹³⁸ generalCao.
• Best documented animal-free result. Chicken cells at 28 million cells/mL over 20 days of repeated harvest, with medium cost cut by 38% (Pasitka et al. 2024, doi:10.1038/s43016-024-01022-w). A companion analysis modelled USD 13.67/kg for a 50:50 cell and plant hybrid product (doi:10.1038/s43016-023-00692-2) ^13,139 generalTrung bình. At 2,000 to 3,000 pg per cell, 28 million cells/mL is about 56 to 84 g/L of wet cells (our arithmetic), below Humbird's 140 to 195 g/L. No comparable data exist for fish or shrimp cells.
• Growth factor cost depends on the price assumption. Humbird models growth factors at only USD 3 to 4 per kg of wet cells at 100,000 t/yr, assuming bulk recombinant prices ¹³⁸ generalCao. At current prices, growth factors can be up to 95% of serum-free medium cost, and medium about 98% of raw-material cost at scale (preprint citing Negulescu et al. 2023) ¹⁴⁰ generalTrung bình. Both are true; they describe different futures (DG-104).
• Production cost models. Humbird: USD 37/kg of wet cell mass in fed-batch (24 x 20 m3 reactors, 6.8 kt/yr) and USD 51/kg in perfusion, with USD 663 million of capital for 6.9 kt/yr ¹⁴¹ generalCao. A 2022 estimate of USD 63/kg at 540 t/yr is known only through a 2026 review ¹⁴² generalThấp.
• Approvals. Singapore lists six cultivated products, including the world's first cultivated duck (April 2026) and beef (July 2026) ¹² VN-adjacentCao. The US FDA has completed five cultivated-cell consultations, including Wildtype's coho salmon on 28 May 2025, the first cultivated seafood cleared in the US ¹⁴³ generalCao. See S3. Regional.

#S13.11.2 Vietnamese evidence

• Vietnam already expresses some of the costliest ingredients of cell culture media, but not for food: FGF-2, KGF/FGF7 and PDGF-BB in Pichia at VNU-HCM ^102,103,104 VN-directTrung bình.
• Vinmec tested four commercial serum-free and xeno-free kits for cord-blood mesenchymal stem cells; only one worked (doi:10.1186/s13287-021-02694-y) ¹⁴⁴ VN-directCao. Other work compares fetal bovine serum with serum-free long-term culture ¹⁴⁵ VN-directCao.
• The only livestock stem-cell paper found is on porcine female germline stem cells (2019) ¹⁴⁶ VN-directCao. The only Vietnamese cultivated-meat paper is a 2024 legal analysis of EU and Singapore rules ¹⁴⁷ VN-directCao.
• We found no fish or shrimp cell line developed in Vietnam, in either database ^1,2 VN-directTrung bình.

#S13.11.3 What it means for Vietnam

• Vietnam has no food-grade fish or shrimp cell line to start from. The Indian catfish lines are a useful research tool for the same species as Vietnamese tra, but would need serum-free adaptation, suspension growth and muscle or fat character ¹³² VN-adjacentCao.
• Any shrimp cultivated-seafood claim built on PmLyO-Sf9 should be treated as misinformed: the line is part insect ¹³⁶ generalCao.
• The science supports "watch, do not build" for food production. Useful public research steps are serum-free adaptation of catfish cells and testing Vietnamese Pichia growth factors on fish cells.

#S13.12 Insects (benchmark only)

Insect protein appears in this report as an incumbent and a benchmark, not as a recommended play.

• Shrimp. Black soldier fly (BSF) larvae meal replaced 20% of fishmeal in whiteleg shrimp without growth loss; 30% reduced growth (control diet 25% fishmeal, so 20% replacement is about 5% of the diet; doi:10.1017/s0007114521004670) ¹⁴ generalCao. BSF meal had lower protein digestibility than the control in a shrimp screen ⁷ generalCao. Yellow mealworm meal replaced 50% of fishmeal with better growth and cut post-challenge mortality by 76.9% (a company-linked product; doi:10.3390/ani9050258) ¹⁴⁸ generalTrung bình.
• Fish. BSF meal replaced up to 60% of fishmeal in juvenile striped catfish (174 g/kg of diet), but 100% cut growth ¹⁴⁹ generalTrung bình. A meta-analysis puts the generally safe limit below 30% of fishmeal ¹⁵⁰ generalTrung bình. Cricket meal replaced 100% of fishmeal protein in snakehead without growth loss ¹⁵¹ generalTrung bình.
• Vietnamese evidence. 30% BSF larvae meal gave the best result in swamp eel (feed conversion ratio 2.33) (doi:10.3390/aquacj5010007); fresh or dried BSF larvae have been tested in snakehead ^152,153 VN-directCao. The BSF cluster (34 Vietnam-tagged records) is led by papers on biodiesel from larval fat ² VN-directTrung bình.
• Industry. Entobel's Vietnamese plants have a design capacity of 11,000 t/yr ^154,155 VN-directTrung bình.
• What it means. As a shrimp-feed benchmark, insect meal displaces only a small share of fishmeal. Microbial proteins such as CAP match or exceed it on digestibility ^7,14 generalCao. Many of the gains claimed for novel proteins in shrimp are functional (survival, palatability) rather than protein supply ^75,148,156 generalTrung bình.

#S13.13 Tropical fermentation

#S13.13.1 State of the science in 2026

• Kluyveromyces marxianus SS106, isolated from Thai cassava pulp and starch-factory soil, grows at 35 to 45 C. On raw cassava starch at 40 C in 5 L stirred tanks it reached 39.3 g/L of cell dry weight (3.28 g/L/h) in a single batch and 36 to 46 g/L in repeated batches. The authors chose 40 C over 42 C to save energy (doi:10.3390/pr8080898) ¹⁵ VN-adjacentCao.
• The genome of K. marxianus DMKU 3-1042 explains its thermotolerance. High-temperature fermentation is "expected" to reduce cooling cost and contamination, but we found no published study that quantifies the cooling cost saved in a tropical plant (doi:10.1186/s13068-015-0227-x) ¹⁵⁷ generalTrung bình.
• K. marxianus is also a precision-fermentation host (leghemoglobin at 7 to 10 g/L), and C. utilis tolerates up to 41 C ^95,9,72 generalCao.

#S13.13.2 Engineering benchmarks

• Aerobic growth releases about 460 kJ per mol of oxygen consumed, and cooling water is 4 to 8 times cheaper than chilled water per unit of heat ¹⁵⁸ generalCao. A German baker's yeast plant with 7 x 150 m3 reactors removes about 11 MW of heat on average ¹⁵⁹ generalCao. A temperate design case uses 0.56 kW of compressor power per ton of refrigeration ¹⁶⁰ generalCao.
• Our indicative cooling penalty for southern Vietnam is about 0.7 to 1.35 kWh per kg of dry biomass, plus chiller capex: about USD 70 to 180 per tonne of dry biomass, or roughly 2 to 4% of the indicative cost of fungal feed protein (1 to 8% at the extremes). This is our calculation from general engineering rules and an assumed design wet-bulb temperature, not a published measurement ^158,160,159 VN-directThấp. Vietnamese manufacturing power costs about USD 0.073 to 0.081 per kWh for a plant running 24 hours a day ^17,161 VN-directCao.

#S13.13.3 Vietnamese evidence

• The Tropical Fermentation Network, set up through the Erasmus+ AsiFood project, links HUST with French and Southeast Asian partners on starters, safety and by-product valorisation (doi:10.3389/fmicb.2018.02278) ¹⁶² VN-directTrung bình.
• We found no Vietnamese measurement of fermentation cooling energy ^157,158 VN-directTrung bình.

#S13.13.4 What it means for Vietnam

• The strain toolkit for 40 to 45 C fermentation exists and matches cassava feedstocks ¹⁵ VN-adjacentCao. Design for thermotolerant strains or hybrid cooling from day one.
• One call to a Vietnamese yeast or MSG plant for its chiller energy per m3 of broth would replace our estimate with a measurement.

#S13.14 AI in research and development

• State of the evidence. This study did not review the literature on artificial intelligence (AI) in protein design, strain engineering or formulation. We make no claim about its state of the art.
• Money. The Bezos Earth Fund's Future of Food programme has committed USD 1 billion by 2030 and lists 23 grants worth USD 194.8 million, several on AI for alternative protein (fungal, cultivated and plant protein); we found no open call ¹⁶³ generalTrung bình.
• Rules in Vietnam. Decision 23/2026 lists omics (item 48) and bioinformatics (item 50) as priority high technologies ¹⁶⁴ VN-directCao. The IP Law amendment (Law 131/2025/QH15, effective 1 April 2026) adds rules on AI-created subject matter and a text and data mining exception, which bear on AI-assisted strain and protein design ¹⁶⁵ VN-directTrung bình.
• Vietnamese capacity. VINIF, Vingroup's research foundation, has a Big Data focus in its project funding ¹⁶⁶ VN-directTrung bình. We found no Vietnamese publication applying AI to alternative-protein R&D.
• What it means. AI tools may shorten strain and formulation development, but we have no evidence in this study to size the effect. Treat AI claims in pitches as unverified until backed by published or audited results.

#S13.15 Corrections to earlier claims

This table collects every science claim from earlier drafts that the wave 2 review checked. Claims marked "Unsupported" or "Wrong" are not used anywhere in the report. Conflicts are logged in R2. Disagreements.

#S13.16 Techno-economic benchmarks used in this appendix

All values are from tea_benchmarks.csv. They are global models, mostly for US or European sites, and must not be read as Vietnamese costs. Capex per annual tonne is our derivation from the verified inputs ^6,70,141 generalCao.

The H2 and CO2 row rests on an abstract only; the others on full text or key excerpts.

#Gaps and how to close them

#Related data files

• science_facts.csv: the 43 checked science claims behind sections S13.3 to S13.13, with DOI, full-text flag and retraction check.
• tea_benchmarks.csv: the 25 techno-economic benchmarks behind section S13.16 (see Costs and benchmarks).
• replacement_trials.csv: the 20 feed replacement trials behind sections S13.5 and S13.12 (see Feed and aquafeed market).
• publications.csv: Vietnam-affiliated publications, including those cited here (see S7. Research landscape).
• biblio_counts.csv: OpenAIRE counts by topic and country.
• cost_inputs.csv: Vietnamese input costs used in the indicative cost estimates.
• disagreements.csv: DG-056, DG-080, DG-087, DG-088, DG-104 to DG-110 and DG-138 to DG-139 (see R2. Disagreements).
• Working papers: working-papers/wave2/science/science.md, working-papers/wave1/costs/costs.md, working-papers/wave1/rnd/m6-rnd-capacity.md, working-papers/wave2/feedmkt/feedmkt.md, working-papers/wave2/biblio/bibliometrics.md.