61 published TEAs, replicated and costed

untangle.bio is a browser tool that draws a bioprocess flowsheet, simulates it and returns the plant’s capital and operating cost. This page asks one question of it: how close do those numbers land to published techno-economic analyses of the same processes? 61 flowsheets were redrawn from their papers or design reports and run unchanged, and each result is set beside the capital, operating cost and cost of goods the source itself published. Fifteen of them have a full case section on this page; the other 46 are rows in the literature section at the end, and every one of the 61 is on the scorecard and in the correlation below.

Read the tiles left to right as a funnel: of the 61 flowsheets, four have no published number to compare against and three more (Succinic acid (glucose), Citric acid and Erythritol) have cost figures with no primary locator and carry no ratio; the 54 that remain are the pairs the correlation and the statistics below are built on, where the like-for-like subset is separated out. Every row is on this page.

The short answer. On the five pairs that compare the same sections of the plant, the same product and the same cost basis, untangle’s cost of goods lands at a median 0.7x of the source’s, and 60% of those pairs sit inside a factor of two. Across all 54 pairs with a traceable published figure, most of which set a narrower untangle flowsheet against a whole plant, the median is 1.4x on cost of goods and 1.2x on capital in 2026 dollars, with 41% and 52% inside a factor of two. The typical pair disagrees by a factor of 2.2x on cost of goods, direction ignored.

What that means for a reader deciding whether to use it. A factor of two either way is the accuracy of an order-of-magnitude estimate: the class of number used to rank routes, size a first plant and decide whether a detailed study is worth commissioning, not the class used to sign a capital budget. On that standard the like-for-like pairs are inside the band and the wider set straddles it, and on every row that runs high the page shows which line of the cost carries the gap. A reader who needs an investment-grade estimate should read this page as evidence of where the model stands, not as a substitute for the study.

52 sources are peer-reviewed papers or preprints; 5 are vendor design cases reconstructed from their published summaries and labelled as such. Every figure taken from a source is traced to a page or table and a status: stated by the source, derived from stated figures with the arithmetic shown, or unverified. A figure without a primary locator is printed but never turned into a ratio. Rows close to the source and rows far from it get the same treatment: the per-operation figures and a reading of what sets the gap.

Protocol

These rules were written down on 2026-09-02, after the first run of the eight new cases and before the runs reported here, and they bind every future run of this page. Where a rule was not followed in the first run, the case notes say so.

1. Case selection. A source qualifies if it is open access or a published vendor design case, states a plant capacity, and reports at least one of total capital, annual operating cost or unit cost for a fermentation or recovery process untangle can draw. Cases were selected by the author of this page; the next tranche should be selected by someone else and held out until the model is frozen.
2. Anchors. Every source figure is recorded in the case file with its value, unit, locator, a short quote and a status: stated, derived by untangle (with the arithmetic), or unverified. Only stated figures, and figures derived from stated figures, may carry a ratio. A derived figure is printed as derived, never as a published one.
3. Feed. The feed is the source's substrate at the source's concentration, at the price the source states or, absent that, the NREL 2011 sugar price. Where the source builds a lignocellulosic front end the harness cannot run, hydrolysate sugar is bought at the price that reproduces the source's own raw-material line (0.30 USD/kg where the source states none) and the case is labelled a lower bound on capital.
4. Fermenter. Titer, yield, host, regime, pH, temperature and batch time are the source's stated values or the midpoint of its stated range. Nothing else on the fermenter is set. Where a file declares a value the source does not state, the declarations list says so.
5. Downstream. Catalogue defaults, with three permitted exceptions, each requiring a note in the file: a zero wash ratio on a membrane the source runs as a clarification, an evaporator concentration factor that reaches the product's solubility, and a crystalliser yield of 0.85 where the source recycles mother liquor. Ion-exchange mode follows the source (capture or flow-through). Every case is also run with these exceptions removed, and both results are shown.
6. Costing. One economic basis for all cases: 2026 USD, the facility grade the source implies, US labour, and an effluent treatment plant sized on the case's own wastewater load. Every case is also costed with that plant excluded, and both results are shown.
7. Comparison. A pair is like-for-like when both sides cover the same sections of the plant, the same product and the same operating-cost basis. The classification is stated on every row, and statistics are reported for all traceable pairs and for like-for-like pairs.
8. One run for every row. No case is run on its own settings: a model change made for any reason is followed by a re-run of every case, and the page is regenerated from the run.

How the numbers are made

untangle prices a flowsheet the same way whether it is drawn on its canvas or run by this benchmark; the benchmark only fixes the inputs. A flowsheet here is a feed stream and an ordered list of unit operations, each with a short list of parameters, plus the economic basis the plant is costed on. This page is generated from the run: every number in every table, figure and sentence resolves from the recorded output and the case files, and the page is rejected if it and the run disagree.

1. The flowsheet file declares the feed composition and its cost, the unit operations in order with any parameter overrides, the facility grade, the selling price and the source’s stated tonnage and anchors with their provenance. It is the auditable object: what was simulated is what the file says, and every override carries a note.
2. Thorough simulation runs the mass balance in mass flow (concentration times volumetric flow), splits each component by rejection or efficiency, and follows the product along the outlet each operation types as its product. Fermenters integrate their kinetics with oxygen transfer, substrate and product inhibition and a carbon balance on the declared yields; dryers, columns, membranes and crystallisers size themselves on the stream they see and refuse a duty they cannot do.
3. Detailed TEA returns untangle’s own capital and operating cost for the flowsheet as simulated. The capital figures on this page are untangle’s internal numbers, reported so they can be set beside the source’s; how they are arrived at is not part of this comparison. Operating cost is built up from raw materials, consumables, utilities, wastewater, labour, QC/QA, maintenance, depreciation and overhead, and the basis of each comparison is printed beside it.
4. Cost of goods divides annual operating cost by the product mass the balance actually delivers at the last step; minimum selling price adds the return on capital.

Scorecard

untangle, all 61 cases

untangle against the sources

Bold ratios are like-for-like. Unbolded ratios are pairs where untangle's scope is narrower than the source's (a lower bound) or where the two sides cost a different product or a different operating-cost basis (marked). Unverified means the source figure exists only in a secondary summary with no primary locator: it is printed in the case section and carries no ratio anywhere on this page. The operating-cost basis column is the source's; untangle's always includes depreciation.

Reading across the 54 rows with a traceable anchor, in the order the statistics support:

• On the median case the ratio sits between 0.6x and 1.4x depending on the measure and the subset (capital, operating cost and cost of goods; all traceable pairs and like-for-like pairs; declared, catalogue-default and no-wastewater runs). The statistics table below prints every one of those medians with its interval and its n.
• Three rows carry no cost ratio at all: Succinic acid (glucose), Citric acid and Erythritol. Their TCI, operating cost and unit cost exist only in secondary summaries of vendor design cases, and one of the unit costs is a division untangle performed itself on two unverified inputs. The vendor's own pages state tonnage, fermentor count and the downstream train, so those rows still support a reading of scale and configuration (five vessels bought against the design case's seven on citric acid), and nothing about cost.
• Lactic acid, stover sits closest to its source on cost of goods among the like-for-like pairs (0.68x is the like-for-like median).
• Lactoferrin is two sources, not one result: its capital figure is a vendor design case with no primary locator (unverified, no ratio), while its operating cost and unit cost (0.34x, 0.34x) are against a 2026 preprint at a different tonnage with its own capital figure.
• Succinic acid from stover (0.50x) and lactic acid (1.06x) on bought sugar are the two whole-plant rows with stated cost figures; on the succinic row the unit cost is the paper's operating cost over its own product mass, derived because the paper publishes a selling price and no unit cost.
• Isobutanol (2.4x on unit cost) is a single distillation step on a 22 g/L broth against the paper's two columns and a decanter; butyric acid (6.6x) compares butyric acid against a source that costs a 30% sodium butyrate solution per tonne of butyrate, so its cost ratio is a basis difference and not a fit statistic.
• Scope and basis comparisons: 2,3-butanediol (a conversion against a plant with free bagasse), the two monoclonal antibodies (like-for-like on the downstream section at 0.69x and 0.66x on downstream operating cost; whole-plant figures are scope), mycoprotein (the source's capital excludes land, working capital and start-up, so it is not a TCI; the source product is wet) and crude enzyme (the source's operating cost excludes capital recovery and its enzyme is a different one).

Correlation with the papers

Each panel plots untangle against the paper on log axes, for the rows with a traceable source figure, with the paper's figure moved to 2026 dollars by the CEPCI ratio so both sides are in one cost year. The diagonal is parity; the shaded band is a factor of two either side. Filled markers are pairs on a like-for-like basis. Hollow markers are pairs where the scope or the basis differs (a downstream-only flowsheet against a whole-plant figure, a plant whose pretreatment section untangle did not build, a different product or operating-cost basis) and are shown for completeness, not for the fit. 22 (Bacteria (generic) (Vlaeminck 2023), Butanol (Carmona-Garcia 2021), Cellulase (Ferreira 2018), Ethanol (Mwanakaba 2025), Ethanol (Wu 2014), GABA (gamma-Aminobutyric Acid) (Jing 2025), Glucose (Quinonez-Ensuncho 2026), Ethanol, Ethanol (Gnansounou 2015), Ethanol (Gubicza 2016), Ethanol (Littlewood 2013), Ethanol (Quintero 2015), Ethanol (Rajendran and Murthy 2017), Ethanol (Srinophakun 2020), Ethanol (Srinophakun 2022), Ethanol (Barta 2010), Levulinic Acid (Meramo-Hurtado 2021), Chicken Myogenic Stem Cells (Humbird 2021), Ethanol (Kautto 2014), Pea Protein (Legumin/Vicilin) (Yang 2025), Thaumatin (Frontiers in Nutrition 2026), Thaumatin (Kelada 2021)) have no stated cost year and are outside the normalised statistics. Hover a marker for the numbers and the basis.

Markers: 1 Lactoferrin · 2 Monoclonal antibody · 3 Plant-based mAb · 4 2,3-Butanediol · 5 Succinic acid, stover · 6 Lactic acid, stover · 7 Isobutanol · 8 Butyric acid · 9 Mycoprotein · 10 Crude enzyme · 11 2,3-Butanediol (Mailaram 2022) · 12 Adipic Acid (Sikazwe 2024) · 13 Bacteria (generic) (Vlaeminck 2023) · 14 Butanol (Carmona-Garcia 2021) · 15 Cellulase (Ferreira 2018) · 16 Coconut water solids (Viana 2026) · 17 Ethanol (Mwanakaba 2025) · 18 Ethanol (Wu 2014) · 19 GABA (gamma-Aminobutyric Acid) (Jing 2025) · 20 Glucose (Quinonez-Ensuncho 2026) · 21 Glycerol (Attarbachi 2024) · 22 Isobutanol (2-Methyl-1-propanol) (Cai 2018) · 23 Polyhydroxyalkanoate (PHA) (Kolonnage 2023) · 24 Succinic Acid (Tran 2023) · 25 Ethanol · 26 Ethanol (Buthelezi 2025) · 27 Ethanol (Gnansounou 2015) · 28 Ethanol (Gubicza 2016) · 29 Ethanol (Huang 2020) · 30 Ethanol (Humbird 2011) · 31 Ethanol (Junqueira 2017) · 32 Ethanol (Kumar and Murthy 2011) · 33 Ethanol (Littlewood 2013) · 34 Ethanol (Muhammad 2020) · 35 Ethanol (Quintero 2015) · 36 Ethanol (Rajendran and Murthy 2017) · 37 Ethanol (Rodrigues Gurgel da Silva 2018) · 38 Ethanol (Silva 2017) · 39 Ethanol (Srinophakun 2020) · 40 Ethanol (Srinophakun 2022) · 41 Ethanol (Barta 2010) · 42 Ethanol (Barta 2010) · 43 Fish protein hydrolysate (He 2015) · 44 Glucose · 45 Lactic Acid (Mailaram 2023) · 46 Levulinic Acid (Meramo-Hurtado 2021) · 47 Acetic Acid (Morales-Vera 2020) · 48 Chicken Myogenic Stem Cells (Humbird 2021) · 49 Ethanol (Kautto 2014) · 50 Lactic acid (Wimble 2025) · 51 Pea Protein (Legumin/Vicilin) (Yang 2025) · 52 RuBisCO (Castro-Dominguez 2026) · 53 Thaumatin (Frontiers in Nutrition 2026) · 54 Thaumatin (Kelada 2021). The same number marks the same case in every panel; a short line joins a number to its marker where the points crowd. Click a number, a marker or a legend entry to open the case.

Statistics

Ratios are untangle divided by the source, with the source in 2026 dollars. Brackets are 95% percentile bootstrap intervals (10,000 resamples, fixed seed 0) on the median and on the share within a factor of two; they are wide because the samples are small, and that is the point of showing them. Median spread is the median of the larger of ratio and its inverse: the factor by which the typical pair disagrees, direction ignored. The last column is the Pearson correlation on log values, which any model that gets tonnage and product class right will score highly on across five orders of magnitude; it is shown because it is conventionally asked for, not because it discriminates between models. Rows whose cost figures have no primary locator (Succinic acid (glucose), Citric acid and Erythritol) are outside every subset, so the n printed on each row is the n used.

In plain terms. On the pairs built the same way on both sides the median is 0.7x and 60% land within a factor of two; on all traceable pairs the median is 1.4x and 41% do, and most of the pairs outside the band are rows where untangle’s flowsheet covers less of the plant than the paper’s figure does.

All figures here are on the 2026-dollar basis. Cost of goods. On the 5 like-for-like pairs the median ratio is 0.7x and 60% of pairs fall within a factor of two. Running the same cases on catalogue defaults instead of the declared downstream overrides gives a median of 0.7x on 5 pairs; excluding the wastewater plant gives 0.7x on 5 pairs. Capital. 0 like-for-like pairs survive the provenance audit (every whole-plant capital figure with a primary locator is either a lower bound because untangle bought the sugar, or not a total capital investment); on all 27 traceable pairs the median is 1.2x and 52% fall within 2x, read as lower bounds. Operating cost. 6 like-for-like pairs, median 0.6x, 50% within 2x; 19 traceable pairs in all. The medians across every measure and subset run from 0.6x to 1.4x, and the intervals on the medians overlap across subsets.

The same statistics without cost-year normalisation

The table above puts every paper in 2026 dollars. This one does not: it is the same arithmetic on the figures exactly as the sources print them, each in its own cost year, which is the basis the scorecard, the wastewater comparison and every case section below use. The difference between the two tables is construction-cost inflation on the untangle side, and nothing else.

Per case, the factor that separates the two. untangle prices in 2026 dollars (CEPCI 808, held at the 2025 average until the 2026 figure is published); a paper's figure is multiplied by CEPCI 2026 over the CEPCI of its cost year, which divides the ratio by the same factor. Where a paper does not state a cost year it is taken as the year before publication and marked as an assumption. CEPCI annual averages: 2007 525.4, 2008 575.4, 2009 521.9, 2010 550.8, 2011 585.7, 2012 584.6, 2013 567.3, 2014 576.1, 2015 556.8, 2016 541.7, 2017 567.5, 2018 603.1, 2019 607.5, 2020 596.2, 2021 708.8, 2022 816.0, 2023 797.9, 2024 795.4, 2025 808.0, 2026 808.0.

The factor is within 2% of one for Lactoferrin, Monoclonal antibody, Mycoprotein, Crude enzyme, Succinic acid (glucose), Erythritol, Adipic Acid (Sikazwe 2024), Coconut water solids (Viana 2026), Glycerol (Attarbachi 2024), Polyhydroxyalkanoate (PHA) (Kolonnage 2023), Ethanol (Buthelezi 2025), Glucose, Lactic acid (Wimble 2025) and RuBisCO (Castro-Dominguez 2026), because the index has been flat since 2022. It matters for the cases costed before the 2022 step in the index: Plant-based mAb, 2,3-Butanediol, Succinic acid (stover), Lactic acid (stover), Isobutanol, Butyric acid, Citric acid, 2,3-Butanediol (Mailaram 2022), Isobutanol (2-Methyl-1-propanol) (Cai 2018), Succinic Acid (Tran 2023), Ethanol (Huang 2020), Ethanol (Humbird 2011), Ethanol (Junqueira 2017), Ethanol (Kumar and Murthy 2011), Ethanol (Muhammad 2020), Ethanol (Rodrigues Gurgel da Silva 2018), Ethanol (Silva 2017), Ethanol (Barta 2010), Fish protein hydrolysate (He 2015), Lactic Acid (Mailaram 2023) and Acetic Acid (Morales-Vera 2020) come down by 12-35% on the normalised basis. Seven of them (Lactic acid, stover, Unit production cost; Lactic acid, stover, Annual operating cost; 2,3-Butanediol (Mailaram 2022), tci usd; 2,3-Butanediol (Mailaram 2022), msp usd per kg; Ethanol (Humbird 2011), tci usd; Ethanol (Muhammad 2020), tci usd; Ethanol (Rodrigues Gurgel da Silva 2018), msp usd per kg) change side of the parity line. Two caveats. CEPCI is a plant-construction index: right for capital, blunt for everything else - raw materials, labour and utilities follow their own indices, which moved less than construction over 2021-2023 - so the normalised operating-cost and cost-of-goods ratios are the largest correction the year difference can justify, not the most likely one. And an assumed cost year is an assumption, not a measured basis.

With and without the wastewater plant

untangle sizes an outside-battery-limits treatment plant on the chemical oxygen demand of every stream that leaves the flowsheet without a recycle drawn, and charges an effluent tariff on it. Most sources do neither: they recycle process water, send effluent to an existing site plant, or treat it off site at a tariff that is not itemised. The table shows every case costed both ways on the same simulation. Excluding the plant lowers capital by 7% and cost of goods by 6% on the median case; the effect is larger on the downstream-only rows, which handle dilute broths and ship little product per litre of effluent (median 12% on cost of goods), and smaller on the whole-plant rows, where the fermenter and its raw materials dominate (6%). On the like-for-like pairs the median cost-of-goods ratio is 0.69x with the plant and 0.68x without it. Neither basis is the right one for every source: a plant that treats its own effluent needs the plant, and a plant on an industrial site with a shared treatment works pays the tariff and not the capital. The declared basis is kept in the statistics because it is what the app returns by default.

Each case is drawn twice at its source value: the large marker is the declared basis with the wastewater plant, the small amber marker the same simulation costed without it, and the segment between them is the shift. Filled large markers are like-for-like pairs. Numbers follow the legend above; click to open the case.

Other papers

The hand-checked cases above are the benchmark. This section is the literature behind them. The library holds 411 other techno-economic papers (every non-pharma paper it contains); 53 of them have a flowsheet written by hand from the paper, run through the same harness and checked against the paper's own figures, one row each below (seven are the cases above, with their own sections). The other 358 could not be modelled - no catalogue route for the core step, no usable cost figure in the paper, or not a bioprocess at all - and are not shown. The 61 cases above also include the pharma cases and the vendor design cases, which are outside this library count, and the pairs here are counted on the paper's own unit cost (its minimum selling price where none is stated), so the n and the medians in this section differ from the statistics above. In all: 22 unit-cost pairs, 26 capital pairs, sixteen operating-cost pairs.

Two flags are computed, not typed. A paper's unit cost outside the p10-p90 band of its sector is a price outlier (eight papers; bands for Biochemicals, Biofuels, Commodities and Food ingredients). A ratio beyond 5x either way is a ratio outlier and is usually a scope mismatch rather than a model result (fifteen rows). A ratio beyond 100x is not a comparison at all; zero such pairs are shown, marked, and kept out of every statistic and chart.

Every comparable unit-cost pair on log axes. Filled markers are whole-plant like-for-like pairs, hollow markers partial comparisons; the band is a factor of two either side of parity.

By sector

Read a paper row the way a case is read above: the ratio means nothing until the scope on both sides is known. Every modelled row has a hand-checked file whose anchors carry a page or table reference and whose comparison block says what is and is not comparable.

All papers

Grouped by sector. Paper unit cost is the paper's own figure in its own cost year - MSP where no unit cost is stated, the assumed selling price where neither is. OUTLIER marks a price outside the sector band; hover for the band. Ratios are untangle over paper on the measures the row's comparison class allows; a dash means no pair. Modelled says whether a hand-checked model produced untangle's numbers; the train underneath is what was run.

What was compared, paper by paper

One entry per paper that carries a comparison reason, a price note or replication notes, each anchor with its page reference.

What the literature says the catalogue lacks

The papers that could not be modelled are a work list, ranked by how many of them each catalogue gap blocks:

1. A gaseous product cannot be a target product. The digester computes its methane but reports it as a vent, so every biomethane, biogas and bio-hydrogen paper measures zero product. A harness change, and the cheapest item here.
2. Transesterification. Every biodiesel paper stops here; the catalogue's ester route is a different reaction.
3. Thermochemical conversion. Gasification with syngas clean-up and Fischer-Tropsch, fast and microwave pyrolysis, hydrothermal liquefaction. Bio-oil and biochar exist as molecules; nothing makes them.
4. Fermentation modes and hosts. Gas fermentation, solid-state fermentation, a cell-recycle inlet on the stirred tank, and an anaerobic yield pair for the two workhorse hosts.
5. Separations. Ethanol dehydration past the azeotrope, a three-phase decanter, supercritical CO2 extraction, oilseed pressing and leaching, a measured partition coefficient on the extractor, biogas upgrading, a wet mill for fibrous biomass.
6. Column behaviour on a heavy volatile. Furfural is classed as a bottoms product and the column boils the whole aqueous feed; a real furfural column takes it overhead as the heteroazeotrope.
7. Electrochemistry. Electrolysis, direct air capture, microbial electrosynthesis, bioelectrochemical cells.
8. Source defects, not model defects. Three PDFs are not the article, one text extract is empty, one paper typesets its results as images. The files record the gap so the rows do not vanish.

Where a paper file exists, its anchors take precedence over the regex-read index, which mis-read a dozen papers (feedstock intake as product output, an input price as a unit cost, a fixed-capital line as total capital, a comparator plant as the paper's own); the details above say which source each row's numbers came from.

Lactoferrin by precision fermentation (vendor design case 2024)

Source: [1] · Scope: Whole plant (fermenter in the flowsheet) · Facility grade: food_grade · Target: 250 t/yr · Selling price used for MSP: 600.00 USD/kg

Train: fed_batch_bioreactor → centrifugation_disc → microfiltration → ultrafiltration_10k → spray_drying

Six staggered 100 m3 fed-batch fermenters on P. pastoris at 100 g/L dry cells and 8 g/L of secreted lactoferrin, then centrifuge, microfiltration, ultrafiltration/diafiltration and a spray dryer: the design case's chromatography-free scenario, step for step. The vendor's page states the tonnage and the train; the titer is our own back-calculation. untangle buys two production vessels and a three-stage seed train, and raw materials are 34% of the operating cost, which is the shape a high-cell-density yeast process has.

The capital sits in the process vessels (6.9 M purchased), the membranes and the wastewater plant (1.1 M purchased on 4,940 L/hr of feed); the design case's 123.0 M TCI is a secondary-summary figure and is printed below without a ratio. The operating cost and unit cost are compared with a different document, the Kaipa 2026 preprint, at 251.3 t/yr and with its own capital; the product-path yield is 66.0%. Agreement on one figure and not the other cannot be read as one result.

Result

Three runs of this case

Separation spec

Capital

Against the paper

Comparability: downgraded from like-for-like: the 1.04x OPEX and 1.05x COGS agreements are against a 2026 preprint, not against the design case whose TCI gives the 1.40x capital ratio.

Anchor status: mixed source, split by figure · Source grade: A_primary_vendor_page for tonnage and the scenario A train; C_secondary_summary for the TCI; B_primary_abstract_only for the OPEX and unit cost · Source operating-cost basis: includes depreciation [inferred] (untangle’s operating cost includes depreciation).

Purchased capital per unit operation

Units is the machine count the estimator bought: parallel units above a per-machine ceiling on the sizing duty shown (hydraulic flow for a flow-sized machine, vessel volume for a batch vessel, bed volume for a packed bed). Duty per unit is that ceiling-split duty.

Operating cost by line

Labour build-up

Itemised raw materials

Carbon closure

Purchased substrate is the sugar charged with the feed plus the fed-batch concentrate; consumed splits into product, biomass, CO2 and by-products and maintenance by the declared yields, and the residual leaves with the broth. Declared against delivered is the yield the file asked for against the yield the balance returned.

Product mass through the train

Mass of Lactoferrin in the outlet that continues downstream, at the matched feed flow. Step recovery is product out divided by product in; a fermenter has no recovery because it makes the product.

Feed as the flowsheet file declares it

pH 6.0, 30.0 °C. Feed cost basis: purchased. The feed flow is iterated until the annual product mass matches the paper’s tonnage, so no part of any gap is a scale artefact.

Fermenter as declared

Reproduce this case on the canvas

Open on the untangle canvas

Or copy the JSON below and paste it onto the untangle canvas (click the canvas, then Ctrl+V or Cmd+V). The same text saved as lactoferrin_flowsheet.json can be dropped on the canvas or opened with Open project. It carries the matched feed flow, every component with the price the run billed, every step with its parameter overrides, the outlet the product followed out of each step, and the economic basis (grade, selling price, hours) the case was costed on. Run Thorough, then Economic Analysis to see the numbers on this page.

{
  "v": 1,
  "products": [
    "Lactoferrin"
  ],
  "feed": {
    "flow_rate": 4990.3,
    "components": [
      {
        "name": "Water",
        "concentration": 850.0,
        "component_type": "water"
      },
      {
        "name": "Glucose",
        "concentration": 150.0,
        "component_type": "sugar",
        "cost_per_kg": 0.4
      }
    ],
    "temperature": 30.0,
    "pH": 6.0
  },
  "steps": [
    {
      "id": "fed_batch_bioreactor",
      "name": "Fed-Batch Bioreactor",
      "handle": "light",
      "parameters": {
        "titer_basis": "specified",
        "product_titer": 8.0,
        "cell_density": 100.0,
        "y_ps": 0.03,
        "organism": "p_pastoris",
        "fermentation_pH": 6.0,
        "fermentation_temperature": 30.0,
        "is_secreted_product": true,
        "oxygen_regime": "aerobic",
        "y_xs": 0.4,
        "max_product_titer_g_l": 12.0,
        "max_fermentation_time_h": 120.0,
        "feed_substrate_g_l": 800.0,
        "aeration_vvm": 1.5,
        "head_pressure_bar": 2.0,
        "oxygen_mole_fraction": 0.6,
        "max_specific_power_kw_m3": 10.0
      }
    },
    {
      "id": "centrifugation_disc",
      "name": "Disc Stack Centrifugation",
      "handle": "light"
    },
    {
      "id": "microfiltration",
      "name": "Microfiltration",
      "handle": "light",
      "parameters": {
        "wash_water_ratio": 0
      }
    },
    {
      "id": "ultrafiltration_10k",
      "name": "Ultrafiltration (10kDa MWCO)",
      "handle": "heavy",
      "parameters": {
        "wash_water_ratio": 2
      }
    },
    {
      "id": "spray_drying",
      "name": "Spray Drying",
      "handle": "heavy"
    }
  ],
  "title": "Lactoferrin by precision fermentation (vendor design case 2024)",
  "basis": {
    "facility_type": "food_grade",
    "product_selling_price_usd_per_kg": 600.0,
    "annual_operating_hours": 8000,
    "labor_profile": "us_food",
    "wastewater_scope": "included"
  }
}

Monoclonal antibody

Source: [2] · Scope: DSP-only (the feed is a clarifier-ready broth; capital and COGS are conversion figures) · Facility grade: pharma_gmp · Target: 2.4 t/yr · Selling price used for MSP: 200,000.00 USD/kg

Train: depth_filtration → affinity_chromatography → viral_inactivation → ion_exchange_cation → ultrafiltration_10k

Downstream only: depth filtration, protein A capture, viral inactivation, cation exchange and ultrafiltration/diafiltration on a 3.0 g/L clarifier-ready CHO broth, at the paper's 2.4 t/yr. The paper publishes its downstream section separately (the resin-platform column of its Table 1, which is the platform the file's packed-bed capture step matches), so its downstream cost per kg and downstream operating cost are like-for-like pairs against this conversion; its whole-plant cost of goods (36,440 USD/kg) and total capital (571.3 M) include the bioreactor suite this flowsheet was not given and are shown as scope comparisons. The paper's cost of goods is an operating-cost figure that ignores capital, while its operating cost carries a facility-dependent line, so the basis is printed beside the pair.

The product path recovers 64.7% against the paper's stated 70.9% downstream yield. On a pharma-GMP grade the support equipment and the DFC/PC factor (8.1 computed against 8.4 tabulated) set the capital, and consumables (51% of operating cost: resin, membranes, buffers) set the cost of goods.

Result

Three runs of this case

Separation spec

Capital

Against the paper

Comparability: single-column now. Previously the economics were read off the membrane column while the capture geometry was read off the resin column; the two differ by under 0.5% on cost but are different processes.

Anchor status: verified, primary full text · Source grade: A_primary_full_text · Source operating-cost basis: excludes capital recovery [inferred] (untangle’s operating cost includes depreciation).

Purchased capital per unit operation

Units is the machine count the estimator bought: parallel units above a per-machine ceiling on the sizing duty shown (hydraulic flow for a flow-sized machine, vessel volume for a batch vessel, bed volume for a packed bed). Duty per unit is that ceiling-split duty.

Operating cost by line

Labour build-up

Itemised raw materials

Carbon closure

No fermenter in this train: the feed is a finished broth and there is no carbon balance to close.

Product mass through the train

Mass of Monoclonal antibody (IgG) in the outlet that continues downstream, at the matched feed flow. Step recovery is product out divided by product in; a fermenter has no recovery because it makes the product.

Feed as the flowsheet file declares it

pH 7.0, 37.0 °C. Feed cost basis: internal_broth. The feed flow is iterated until the annual product mass matches the paper’s tonnage, so no part of any gap is a scale artefact.

Reproduce this case on the canvas

Open on the untangle canvas

Or copy the JSON below and paste it onto the untangle canvas (click the canvas, then Ctrl+V or Cmd+V). The same text saved as mab_flowsheet.json can be dropped on the canvas or opened with Open project. It carries the matched feed flow, every component with the price the run billed, every step with its parameter overrides, the outlet the product followed out of each step, and the economic basis (grade, selling price, hours) the case was costed on. Run Thorough, then Economic Analysis to see the numbers on this page.

{
  "v": 1,
  "products": [
    "Monoclonal antibody (IgG)"
  ],
  "feed": {
    "flow_rate": 157.8,
    "components": [
      {
        "name": "Water",
        "concentration": 977.0,
        "component_type": "water"
      },
      {
        "name": "Glucose",
        "concentration": 5.0,
        "component_type": "sugar",
        "cost_per_kg": 0.4
      },
      {
        "name": "CHO cells",
        "concentration": 15.0,
        "component_type": "cell",
        "is_suspended": true
      },
      {
        "name": "Monoclonal antibody (IgG)",
        "concentration": 3.0,
        "component_type": "protein"
      }
    ],
    "temperature": 37.0,
    "pH": 7.0
  },
  "steps": [
    {
      "id": "depth_filtration",
      "name": "Depth Filtration",
      "handle": "light"
    },
    {
      "id": "affinity_chromatography",
      "name": "Affinity Chromatography",
      "handle": "light"
    },
    {
      "id": "viral_inactivation",
      "name": "Viral Inactivation",
      "handle": "light"
    },
    {
      "id": "ion_exchange_cation",
      "name": "Cation Exchange Chromatography",
      "handle": "light"
    },
    {
      "id": "ultrafiltration_10k",
      "name": "Ultrafiltration (10kDa MWCO)",
      "handle": "heavy"
    }
  ],
  "title": "Monoclonal antibody",
  "basis": {
    "facility_type": "pharma_gmp",
    "product_selling_price_usd_per_kg": 200000.0,
    "annual_operating_hours": 8000,
    "labor_profile": "us_food",
    "wastewater_scope": "included"
  }
}

Plant-based monoclonal antibody

Source: [3] · Scope: DSP-only (the feed is a clarifier-ready broth; capital and COGS are conversion figures) · Facility grade: pharma_gmp · Target: 0.3 t/yr · Selling price used for MSP: 121,000.00 USD/kg

Train: filter_press → depth_filtration → ultrafiltration_10k → affinity_chromatography → viral_inactivation → ion_exchange_cation → membrane_chromatography → ultrafiltration_10k

Re-wired on 2026-09-02 to the paper's own train: press filter and depth filter for clarification, tenfold ultrafiltration, protein A capture, viral inactivation, cation exchange, a membrane-chromatography polish and a final ultrafiltration/diafiltration, on a 0.5 g/L plant extract carrying 8.0 g/L of solids into the first filter. The old file sent unclarified extract to protein A and polished with an anion exchanger; the simulator refused both and shipped 12.4%. The product path now recovers 67.2% against the paper's stated 65%, with 22% lost at the affinity step and 7% at the cation exchanger.

Nandi 2016 is the one source in this set that publishes its cost of goods on both operating-cost bases: 121,000 USD/kg with depreciation and 90,000 USD/kg without, so depreciation is 26% of its figure. untangle's depreciation is 13% of its operating cost on this row, and its cash-basis cost of goods is compared with the paper's figure without depreciation in the table below. The whole-plant figures include plant growth and extraction that this DSP-only flowsheet was not given, so only the downstream operating cost is a like-for-like pair. To reach the paper's 0.3 t/yr the harness pushes 112 L/hr of extract through GMP columns sized for it, which is why consumables are 1.4 M/yr.

Result

Three runs of this case

Separation spec

Capital

Against the paper

Comparability: anchored for the first time. The row was unanchored only because the flowsheet was mis-wired; that was fixed on 2026-09-02 (see replication_notes), so the paper's Base Case figures are now usable. This case is the literature's only source publishing a cost of goods on BOTH OPEX bases.

Anchor status: verified, primary full text (anchored 2026-09-02) · Source grade: A_primary_full_text · Source operating-cost basis: both bases published (untangle’s operating cost includes depreciation).

Purchased capital per unit operation

Units is the machine count the estimator bought: parallel units above a per-machine ceiling on the sizing duty shown (hydraulic flow for a flow-sized machine, vessel volume for a batch vessel, bed volume for a packed bed). Duty per unit is that ceiling-split duty.

Operating cost by line

Labour build-up

Itemised raw materials

Carbon closure

No fermenter in this train: the feed is a finished broth and there is no carbon balance to close.

Product mass through the train

Mass of Monoclonal Antibody (IgG) in the outlet that continues downstream, at the matched feed flow. Step recovery is product out divided by product in; a fermenter has no recovery because it makes the product.

Feed as the flowsheet file declares it

pH 7.0, 25.0 °C. Feed cost basis: internal_broth. The feed flow is iterated until the annual product mass matches the paper’s tonnage, so no part of any gap is a scale artefact.

Reproduce this case on the canvas

Open on the untangle canvas

Or copy the JSON below and paste it onto the untangle canvas (click the canvas, then Ctrl+V or Cmd+V). The same text saved as plant_mab_flowsheet.json can be dropped on the canvas or opened with Open project. It carries the matched feed flow, every component with the price the run billed, every step with its parameter overrides, the outlet the product followed out of each step, and the economic basis (grade, selling price, hours) the case was costed on. Run Thorough, then Economic Analysis to see the numbers on this page.

{
  "v": 1,
  "products": [
    "Monoclonal Antibody (IgG)"
  ],
  "feed": {
    "flow_rate": 124.1,
    "components": [
      {
        "name": "Water",
        "concentration": 988.5,
        "component_type": "water"
      },
      {
        "name": "Glucose",
        "concentration": 3.0,
        "component_type": "sugar",
        "cost_per_kg": 0.4
      },
      {
        "name": "Fungi/Mold",
        "concentration": 8.0,
        "component_type": "cell",
        "is_suspended": true
      },
      {
        "name": "Monoclonal Antibody (IgG)",
        "concentration": 0.5,
        "component_type": "protein"
      }
    ],
    "temperature": 25.0,
    "pH": 7.0
  },
  "steps": [
    {
      "id": "filter_press",
      "name": "Filter Press (Plate & Frame)",
      "handle": "light"
    },
    {
      "id": "depth_filtration",
      "name": "Depth Filtration",
      "handle": "light"
    },
    {
      "id": "ultrafiltration_10k",
      "name": "Ultrafiltration (10kDa MWCO)",
      "handle": "heavy",
      "parameters": {
        "concentration_factor": 10,
        "wash_water_ratio": 0
      }
    },
    {
      "id": "affinity_chromatography",
      "name": "Affinity Chromatography",
      "handle": "light"
    },
    {
      "id": "viral_inactivation",
      "name": "Viral Inactivation",
      "handle": "light"
    },
    {
      "id": "ion_exchange_cation",
      "name": "Cation Exchange Chromatography",
      "handle": "light"
    },
    {
      "id": "membrane_chromatography",
      "name": "Membrane Chromatography",
      "handle": "light"
    },
    {
      "id": "ultrafiltration_10k",
      "name": "Ultrafiltration (10kDa MWCO)",
      "handle": "heavy",
      "parameters": {
        "concentration_factor": 5,
        "wash_water_ratio": 2
      }
    }
  ],
  "title": "Plant-based monoclonal antibody",
  "basis": {
    "facility_type": "pharma_gmp",
    "product_selling_price_usd_per_kg": 121000.0,
    "annual_operating_hours": 8000,
    "labor_profile": "us_food",
    "wastewater_scope": "included"
  }
}

2,3-Butanediol

Source: [4] · Scope: DSP-only (the feed is a clarifier-ready broth; capital and COGS are conversion figures) · Facility grade: chemical · Target: 2,699 t/yr · Selling price used for MSP: 3.00 USD/kg

Train: centrifugation_disc → microfiltration → thin_film_evaporator → distillation

A four-step recovery train (disc stack, microfiltration, thin-film evaporator, distillation) costed as a conversion. The 22.3 M DSP-only capital is 11x the paper's own 1.95 M DSP figure, and that gap is two things: a 0.98 M purchased wastewater plant installed at its own factor, which the paper does not build, and a TCI-to-purchased-equipment multiplier of 5.70 on a short train (the DFC/PC factor alone is 4.16). Against the paper's whole-plant 22.4 M it is 1.00x, a lower bound that is already above.

Depreciation and maintenance are 38% of the operating cost, so the 2.5x on unit cost is the capital gap carried through, not a utilities or consumables disagreement. The paper's declared scenario gets its bagasse free and most utilities from the host sugar mill, so its operating cost sits structurally below any plant that buys them; the paper's scenario 2 (bagasse bought) is the closer comparator and is recorded in the case file. The minimum selling price anchor is the declared scenario's 2.37 USD/kg.

Result

Three runs of this case

Separation spec

Capital

Against the paper

Comparability: the file is dsp_only against a whole-plant paper, so capital is a lower bound. The declared scenario 1 gets its feedstock and most utilities free, which makes its OPEX and unit cost structurally lower than any plant that buys them.

Anchor status: verified, primary full text · Source grade: A_primary_full_text · Source operating-cost basis: includes depreciation (untangle’s operating cost includes depreciation).