REGAL showed a survival benefit, though SELLAS has yet to release the full data. From what’s public, the benefit could be much smaller than the topline suggests. We get $5.48 a share if SELLAS launches GPS itself and $9.20 if a buyer takes over. Both are below the $13.00 offering price and the $17.95 close. Our case that gets above $17.95 assumes more eligible patients, higher uptake, and better odds of approval.
SELLAS Life Sciences (Nasdaq: SLS) had roughly halved from its August high by the time its Phase 3 REGAL trial read out. More than a quarter of its shares were sold short at the last reported count. Our read is that investors were cautious at best and probably leaning toward a miss. The win was a surprise. So was its size. The control arm lived far longer than we had penciled in, and patients on galinpepimut-S (GPS), a WT1 peptide vaccine, still reached a median overall survival (OS) of 24 months against just under 15 on investigator-chosen best available therapy (BAT). SELLAS gained nearly $2.4B in market value that day, then priced a $150M offering at $13.00, 28% below the next day’s $17.95 close.
The commercial catch is how few patients get to the point where GPS is an option. Most patients with acute myeloid leukemia (AML) who relapse never get back into remission. Of those who do, some go on to transplant. For the patients who go without one, a maintenance drug that extends survival would fill a real gap. There just aren’t many of them.
We can get to the $13 deal price on a buyer’s math, but only with better uptake and more time on treatment than we model. The $17.95 close is harder to get to. Even if a buyer can charge more and keep patients on GPS longer, it still needs to reach nearly nine in ten eligible US patients in our central case, and that is with generous credit for the rest of the pipeline. On the same basis as our own value, it needs more patients than there are. We’d want to see the full survival curves before paying for a longer course, and the launch before paying for that kind of share.
The win clears the bar, with little to spare
Where GPS fits. Most adults with AML get intensive chemotherapy or venetoclax plus a hypomethylating agent first, and many reach remission. Most of those relapse. A minority of relapsed patients reach a second remission with salvage treatment, and the fit ones with a donor go to a stem-cell transplant, the only treatment that cures many of them. REGAL enrolled the rest: patients in second or later remission who weren't going to transplant, for whom the usual options are observation or continued low-intensity treatment until relapse. GPS is meant to fill that gap as maintenance, an injection every few weeks to train the immune system against leftover leukemia cells.
The bar moved during the trial. The original design, shown at SELLAS's 2019 investor day, assumed GPS would nearly double median survival, 10.4 vs 5.4 months (an HR of 0.52), with the final analysis at 105 deaths. Deaths came more slowly than planned, and a 2022 design update cut the final analysis to 80. That raised the bar: an observed HR of ~0.636 or better now counted as significant, which SELLAS translated as 12.6 vs 8.0 months. The published protocol claims at least 90% power at that HR. Our calculation, with the protocol's one-sided 2.5% test and O'Brien-Fleming boundary, gives ~50% power at 80 deaths if the true HR is 0.636. So 0.636 is the bar to clear, not an expected effect. The observed 0.59 clears it.
The two headline numbers fit the protocol's own analysis. At first sight an HR of 0.59 with p=0.035 looks odd at 80 deaths, and it drew plenty of comment on X. We simulated REGAL-sized trials (126 patients, 1:1 randomization, analysis at the 80th death, medians of 24.0 and 14.7 months) under three survival-curve shapes. Analysed the simple way, trials that land on an HR near 0.59 give a median p of ~0.020. Analysed the way the protocol specifies, with a Cox model stratified by REGAL's four randomization factors, the median is ~0.033, almost exactly what SELLAS reported (Exhibit 1). Four factors make up to 16 groups in 126 patients, so each holds only a handful of deaths. In our model that costs ~15% of the statistical information: 80 deaths behave like ~67. On that much information the bar tightens from ~0.64 to ~0.61, and 0.59 still clears it.
So the pair isn't a red flag, but the interval is wide. Back-calculating from the HR and p-value puts the undisclosed 95% confidence interval at ~0.36-0.96, and our simulations give much the same (~0.36-0.97). The top of that range sits close to no effect, so the true benefit could be a good deal smaller than the headline. Estimates that only just clear a significance bar also tend to overstate the true effect. A smaller analysed population reproduces both reported numbers in 4.3% of our simulations, against 6.2% for all randomized patients, which is why the analysis set still matters. Only the full data can confirm the test, the interval and who was counted.
REGAL's HR and p-value fit the protocol's stratified analysis
The reported p-value is consistent with the analysis the protocol specifies. Among simulated trials landing on an HR of 0.57-0.61, the stratified analysis gives a median p of 0.033 against a reported 0.035; the simple unstratified analysis gives 0.020.
The rest of the disclosure is thin too. Exhibit 2 sets out what's known and what isn't. There is no confidence interval, no analysis set, no per-arm deaths, no Kaplan-Meier curves, no transplant counts and no BAT mix. None appear in the release, on the webcast slide or in the offering prospectus. The published protocol specifies a stratified Cox analysis in all randomized patients, and our simulations suggest that is what was reported, though only the full data can confirm it. The registry also shows where REGAL ran: only 17 of 73 listed sites are in the US. The ~18-minute webcast was a prepared statement with no Q&A, on the same day the company launched its financing.
What SELLAS has and has not disclosed about REGAL
The topline tells us GPS met its endpoint. It doesn't yet tell us how large and durable the benefit is, or which patients drove it.
| Item | Disclosed? | Comment |
|---|---|---|
| Primary endpoint met (OS) | Yes: HR 0.59, p=0.035 | On the release and the webcast slide |
| Median OS by arm | Yes: 24.0 vs 14.7 months | |
| Deaths at final analysis | Yes: 80 | |
| Confidence interval for the HR | No | Not in the release, webcast slide or prospectus. Our simulation implies ~0.36-0.97 |
| Analysis set and N analyzed | No | Protocol says stratified Cox in all randomized patients. Registry lists 127 enrolled vs 126 randomized |
| Deaths and censoring per arm; KM curves | No | Needed to judge the shape of the benefit |
| Transplant after randomization | Partly: "limited", balanced | No counts |
| Where patients were treated | Partly | 17 of 73 listed registry sites are in the US (46 Europe, 5 India, 5 Taiwan). Patient split not disclosed |
| Control-arm mix, subgroups, secondary endpoints | No | Full presentation planned at a medical meeting |
| Q&A on the readout webcast | No | ~18-minute prepared statement; offering launched the same day |
Bottom line: without the confidence interval, Kaplan-Meier curves and deaths by arm, the size and durability of the survival benefit can't yet be judged.
None of this means the result is wrong. It does mean the market added $2.4B of value on three numbers from an open-label trial, before anyone saw the curves.
How to read the full REGAL data. Four disclosures decide whether the topline holds and what it is worth:
- Who was analyzed. The protocol specifies all 126 randomized patients. A smaller analysis set reproduces both reported numbers in 4.3% of our simulations, against 6.2% for all randomized patients, so the topline alone can't tell them apart.
- The confidence interval. We back-calculate ~0.36-0.96. An upper bound that close to 1 would mean the true benefit could be much smaller than an HR of 0.59 suggests.
- The shape of the curves. A tail of patients still in remission lengthens the course: each year past year two adds four doses, ~$88k per patient at our peak price. A tail of patients who have stopped GPS adds survival but no revenue.
- Control treatment and transplants by region. 25 of the 73 listed sites are in Greece, Poland, Hungary and Serbia. How many control patients got venetoclax or only observation, and how many in each arm went to transplant, decides how far the 14.7 months travels.
The control arm rewrote the benchmark
The most telling number in the release is 14.7 months. SELLAS's 2022 design update used 8.0 months as its benchmark, and management itself cited ~8-10 months as the real-world norm. GPS reached 24.0 months, also above the 12.6 months SELLAS used to illustrate the significance bar. Our own pre-readout work assumed ~8 months for the control arm and got it wrong. Exhibit 3 shows how far off the field was. Every second-remission benchmark we could find sits at 5-8 months, including the Moffitt historical control behind SELLAS's earlier pilot. REGAL's control arm instead matched the placebo arm of QUAZAR (14.8 months), a first-remission trial. GPS, at 24.0 months, sits right beside oral azacitidine's 24.7 in that same trial.
REGAL's control arm outlived nearly every benchmark
The control arm reached 14.7 months against an 8.0-month design benchmark. That is level with QUAZAR's first-remission placebo arm (14.8). GPS reached 24.0, beside oral azacitidine's 24.7 in that trial. Only a selected cohort of venetoclax responders (Padua, 20.1 months) lived longer.
Patient selection probably explains a lot of this. REGAL enrolled patients with marrow blasts under 5%, recovered blood counts, and a life expectancy over six months. "Transplant-ineligible" also covered patients who chose not to transplant or had no donor, not just frail ones. WT1 screening was dropped after the first 20 patients, because none failed it. That is a favorable window for a vaccine: little disease to control and a marrow that has recovered. It also means REGAL's absolute survival numbers may not carry over to a broader, sicker treated population. The 9.3-month gain in medians is the better guide. The 24 months is not a number to plug into a market model.
The pace of deaths is consistent with a long tail. SELLAS reported the 60th death in December 2024, the 72nd a year later and the 80th only in late October 2026: eight deaths in ten months. Counting from the 60th death, with 66 patients still alive and each arm's death rate held at the level its reported median implies, the last 20 deaths should have taken about ten months, putting the 80th around October 2025, a year before it came. In our simulations, models with a group of long-term survivors match the actual dates about three times as often as models without one; heavier loss to follow-up than usual also helps. For a vaccine that is plausible: an immune response takes months to build, and the class tends to show late, flat tails rather than early separation. Favorable-risk cytogenetics would produce a tail too, since those patients can live for years without a transplant. Curves by arm, with numbers at risk and censoring, will show where the tail sits; baseline risk by arm will show why. It matters for value as well as for the evidence. GPS is given until relapse, so a long tail means a long course only if it is made of patients still in remission.
About 1,100 US patients a year could start GPS, and maybe fewer
REGAL's population is narrow by design. The funnel in Exhibit 4 starts from 22,720 projected US AML cases in 2026 (ACS) and splits by age, because remission, relapse and transplant rates all shift with it. Under-60s mostly reach remission and, when they relapse and recover, usually go to transplant. Over-75s are often untreated or never reach a second remission. The 60-74 band supplies about half the pool. We end at ~1,115 patients a year who reach second remission, aren't transplanted and are fit to start maintenance, with a range of ~420-2,100. We add ~15% for patients in third or later remission, since a REGAL-based indication would cover them.
About 1,100 US patients a year could start GPS maintenance
An age-stratified funnel from new US AML cases to patients who could start GPS. It starts from 22,720 cases. About half the pool (544 of ~1,115) comes from 60-74 year-olds: younger patients mostly go to transplant, and few over 75 reach a second remission.
Patient-level data back the age split. In Beat AML, a public dataset of 700 adults with AML (excluding the APL subtype) treated at eight academic centers, mostly before venetoclax, 58% of 18-59 year-olds had intensive 7+3 chemotherapy and never a hypomethylating agent, and 43% went to transplant at some point. Among over-75s, 59% had a hypomethylating agent and never 7+3, and 1% were transplanted. The first-remission rate at 60-74 (58%) matches our 60%. Under 60 it was 68%, below our 75%, which points the pool a few percent lower. The dataset can't size the second-remission pool itself, because it has no remission dates or transplant-eligibility fields.
SELLAS's own numbers agree on the order of magnitude. Its investor deck puts second remission at 12-15% of AML patients, a figure adapted from a 2010 Japanese cohort. Applied to US incidence, that is 2,726-3,408 patients reaching second remission, against our 2,546 at the same step. The source cohort was younger than US patients, and younger patients reach second remission more often, so we treat it as a ceiling.
The 10-K tells a bigger story. SELLAS's 10-K counts ~8,750 eligible patients across the US, Europe's four largest markets and Japan, ~3,750 of them in the US. It gets there by assuming half of all second-line patients reach a second remission, and cites two papers for it. Neither supports the number. One reports a 21.6% complete-remission rate for ivosidenib in IDH1-mutated relapsed AML; the other found second remissions in 4 of 11 patients who relapsed more than five years out. The 10-K count also skips the transplant filter REGAL applied at entry. Exhibit A1 in the appendix walks from 3,750 to our ~960. The gap matters: the 10-K's number clears every patient requirement in Exhibit 5, and ours clears only the most generous.
The venetoclax era probably shrinks the pool further. Most patients aged 60-74 who can't take intensive chemotherapy now get venetoclax plus a hypomethylating agent first, and stay on it continuously. When they relapse, salvage can still produce remissions, but they rarely last. In MD Anderson's series of 41 such patients, median survival was under three months. In Mayo's 2026 series of 356, about half of the 99 given venetoclax-based salvage reached a complete remission (with or without full count recovery), yet their median survival was only 6.0 months. Those who do reach a second remission usually stay on whatever got them there, so there is no clean window for a standalone maintenance drug. Cutting the 60-74 band by ~25-40% for this takes the total pool to ~960 a year; we treat that as a downside case, not the base. REGAL's own design sharpens the point. The protocol let the BAT arm continue venetoclax, though how many did is undisclosed. Patients who reached second remission on venetoclax and a hypomethylating agent could enroll only if side effects were delaying their cycles, and those randomized to GPS had to stop that regimen. A label built on REGAL asks doctors to take patients off treatment that is working, which is a barrier to switching, not an opening.
The one approved AML maintenance drug sets a sober pace. Oral azacitidine (Onureg) has a larger first-remission label and an OS benefit. It sold $95M in the US in its second full year, 2022, and $117M in 2023, or ~420 sales-implied patient-years at an estimated ~$281k net a year. Bristol Myers Squibb stopped breaking out its sales from 2024. Our GPS base case already assumes better uptake than that: 50% of eligible patients, or ~640 treated in the US a year. The base valuation runs on the 1,115-patient funnel plus later remissions (~1,280); on the evidence-adjusted ~960, or ~1,100 with later remissions added, the same 50% share gives ~550. The base already sits above that case.
What $17.95 requires
We back-solve the price rather than set a target. We start from the $3.66B pre-offering market cap, subtract cash, and give generous credit to everything except the REGAL label: $450M for SLS009 and $200M for solid tumors, both above our own values in the next section, plus the first-remission option. What's left has to come from REGAL. We price GPS per injection on the trial registry's schedule: 15 doses in year one, every two months in year two and every three months after that, until relapse. That puts revenue at ~$397k per patient started, assuming $300k for the first year at launch, rising 2.5% a year, and 1.6 years on therapy. Ex-US adds 45%, since Europe's five largest markets plus Japan see about as many AML cases as the US.
At $17.95, SELLAS needs more patients than second remission supplies
With generous credit for the rest of the pipeline, only an acquirer with a long course fits at $17.95, at ~87% of eligible patients. On the same basis as our $9.20 value, nothing fits at $17.95, and even $13 needs 65-91% of eligible patients on acquirer economics.
Courses use per-injection pricing on the registry schedule: base $300k first year and 1.6 years on therapy; long course $375k and 2 years. Ex-US at 45% of US; supply includes the +15% later-remission add-on. Modeled, not observed.
Even on generous credit, only one combination fits at $17.95 (Exhibit 5). Exhibit 5 is a deliberately generous screen: it credits the rest of the pipeline at more than we value it and uses the pre-deal share count. On that basis a self-launch at our base course needs ~3,230 US patients treated a year, more than everyone eligible even in our high funnel. On an acquirer's economics it needs ~1,540 at the base course. With a longer, pricier course (~2 years at ~$375k for year one) it needs ~1,110. Only that last case fits under our central ceiling of ~1,280, and only by treating ~87% of eligible patients. The high funnel (~2,430) would accommodate more combinations, but it assumes every ratio at the favorable end.
On the same basis as our $9.20 value, even $13 is a stretch. With ~229M diluted shares, post-deal cash and our own values for the rest of the pipeline, an acquirer needs ~1,850 patients a year at the base course and ~1,330 on the long course to justify $17.95. At the $13.00 offering price it still needs ~1,160 and ~840: 65-91% of our central ceiling. A self-launch needs ~1,940-2,700 at $13. Even with approval certain, our simulated acquirer value has a median of $10.44, ~20% below $13. So the deal price already asks a buyer to treat most eligible patients.
The course length is a sensitivity. The reference case uses 18 doses per patient, a patient who stays on GPS for 1.6 years. The average across everyone who starts could be lower, because patients who stop early miss the frequent first-year doses. At 16 doses, revenue per start falls from ~$397k to ~$353k and modeled acquirer value falls from $9.20 to ~$8.68, holding other inputs fixed.
So why is the stock at $18? Because the price implies different inputs from ours. Our simulation drew 100,000 combinations of patient pool, share, price, treatment length, ex-US sales and SLS009's readout. Share, price and duration moved together (Exhibit 6). The median came out at $5.55 for a self-launch and $9.32 for an acquirer. Acquirer value cleared $17.95 in 0.3% of draws, and in 2.1% assuming GPS is approved. It cleared the $13.00 deal price in ~10%, ~21% if GPS is approved. A favorable commercial case does reach $19.87, on these inputs:
- ~1,500 eligible patients a year, plus a 20% later-remission add-on (~1,800 in all)
- 65% share
- $375k for the first year
- 2 years on therapy
- ex-US sales at 60% of the US, a 10% discount rate and peak sales held to 2041
- 90% approval odds
- a 65% margin
Three of those look aggressive to us. On patients, our evidence check points nearer 960. On share, Onureg's sales imply ~420 US patient-years on drug in its third full year; this case implies ~2,300 a year on GPS (~1,170 new patients, about two years each). On approval odds, 90% is a lot for a single open-label trial, with most listed sites outside the US, whose confidence interval likely reaches close to 1. With our house settings for the add-on, ex-US sales, discount rate and exclusivity (15%, 45%, 9% and 2040), the same case gives ~$18.73, ~4% above the close.
Acquirer value tops $17.95 in under 1% of 100,000 scenarios
Our model clears the close only at the favorable end of its input ranges. Acquirer value tops the $13.00 offering price in ~10% of scenarios, ~21% if GPS is approved. A favorable commercial case (more patients, higher share, price and ex-US sales, 90% approval odds) reaches $19.87, ~11% above the close.
Share, price, duration and funnel are correlated (rho 0.5). SLS009 frontline drawn as pass (40%) or fail. ~229M diluted shares. Ticks on the axis mark the medians. Shares of scenarios depend on the input ranges chosen and are not probabilities. Modeled, not observed.
Management says it will "evaluate every path", and with 14 employees a sale or partnership is the natural route to launch. The stock may be pricing in a takeover, but nothing disclosed shows buyer interest. On our numbers a buyer's economics don't reach even the $13.00 deal price without treating most eligible patients. Paying $18 would mean paying today for assumptions that the full data, the FDA and the first year of launch still have to prove.
The rest of the pipeline is early, and most of its trials are completed
At ~$3.7B investors are buying more than one label, so we valued every program. SELLAS's pipeline chart says a lot on its own. Only two programs are active: REGAL, and SLS009 in high-risk newly diagnosed AML. Everything else is marked completed. That includes all four GPS solid-tumor studies (mesothelioma with nivolumab and alone, ovarian cancer with pembrolizumab and with nivolumab). It also covers SLS009 in relapsed AML, relapsed peripheral T-cell lymphoma (PTCL) and other blood cancers, plus an NIH pediatric program. Management highlights WT1 expression in ~20 tumor types, but no solid-tumor trial is running. The best data are a 10-patient mesothelioma Ph1 with nivolumab (median OS 13.5-17.6 months, single-arm) and a 7.7% response rate in ovarian cancer. We give that an 8% chance of reaching market.
SLS009 (tambiciclib), a CDK9 inhibitor, is the real second asset, and its next readout is close. In patients whose AML had failed venetoclax, adding SLS009 to venetoclax and azacitidine produced responses in 20 of 59 evaluable patients (34%) across doses (ASH abstract). Eight of those 20 were leukemia-free marrows without count recovery; complete remissions, with or without full count recovery, were 12 of 59 (20%). The higher rates SELLAS highlights, 54% in monocytic and 57% in TP53-mutated AML, come from the selected 30 mg twice-weekly dose; across all doses they were 44% and 36%. Median survival of 5.8-8.8 months compares with a historical ~2.5, but Mayo's 2026 series found 6.0 months in patients given venetoclax-based salvage after the same failure. These are single-arm results, and the headline subgroup rates come from the dose SELLAS went on to select. The frontline Ph2 (~80 patients) targets patients predicted not to respond to venetoclax and azacitidine, plus early non-responders. SELLAS had enrolled 40 of them by mid-November and guides to preliminary data in Q4, without saying which cohorts, arms or how much follow-up. The trial is randomized: the public registry lists venetoclax-azacitidine control arms in both frontline groups, so the Q4 data may include an in-trial comparison. Judge any Q4 number against the control arm in the same cohort, and look at early deaths and infections as closely as at remissions. Adding a third drug to that backbone has gone wrong before: in ENHANCE-3, magrolimab added no survival benefit and more fatal adverse events. We treat the readout as a coin flip leaning negative: a 40% chance it supports registration, then 35% odds of approval and a $600M peak. In our model a fully supportive result adds ~$0.85 a share on self-launch economics (~$1.53 for an acquirer), and losing the frontline opportunity costs ~$0.56 (~$1.02). In our simulation, that one readout is among the biggest swing factors in SELLAS's value.
REGAL carries ~59% of SELLAS's pipeline value
Only the REGAL label pairs high odds with meaningful value. Everything else launches in 2030 or later at 8-40% odds and adds ~$0.4B. On either set of economics, value per share sits closer to the pre-readout close than to the $17.95 close.
REGAL is ~59% of the pipeline's value: ~$538M of ~$917M on self-launch economics. Everything else launches in 2030 or later at 8-40% odds. The whole pipeline is worth ~$0.9B on self-launch economics and ~$1.8B on an acquirer's. With cash and the Greater China partnership, that gives $5.48 and ~$9.20 a share. The China licence is intact after an arbitration ended in July 2026, with up to $191.5M of milestones still possible (8-K).
How a buyer's math differs from SELLAS's. The acquirer case changes three inputs, each for a named reason:
- Margin, 60% vs 40%. A buyer already has a hematology sales force and payer contracts, so launch costs are shared.
- Discount rate, 9% vs 11%. Our assumption for a large drugmaker's cost of capital against a single-asset small cap's.
- Approval odds, 80% vs 70%. The clinical data are the same whoever owns GPS, so this gap rests on execution only. GPS is a four-peptide vaccine mixed with the adjuvant Montanide and given alongside GM-CSF injections. SELLAS has never had a product approved, and it buys GM-CSF and Montanide from single suppliers without supply agreements (10-K). FDA did endorse its manufacturing plan at a 2023 meeting, which narrows that risk without removing it. A large buyer can also fund another study if FDA asks for one, which SELLAS could not easily do.
With approval odds at 70% for both owners, acquirer value falls from $9.20 to ~$8.62.
Those values flatter SELLAS. They don't deduct what it owes its licensors: a tiered mid-single-digit royalty to Memorial Sloan Kettering on GPS, plus up to $17.4M in milestones per product, and low-to-high single-digit royalties to GenFleet on SLS009 (10-K). At ~5% of sales, royalties alone would take ~$0.5 a share off the self-launch value and ~$0.6 off the acquirer's. Nor does the model deduct SELLAS's own costs before and around launch, such as the BLA, manufacturing scale-up and a sales force; it applies a margin to sales only. Every $100M of present-value cost outside those margins takes ~$0.44 a share off either value. A self-launch also books ex-US sales at the full 40% margin. A 14-person company would more likely partner them, and halving the ex-US value costs ~$0.36.
Exclusivity matters more than the patent headlines. The GPS composition patents on the four peptides run to 2031-33. The often-quoted 2040 patents cover the seven-peptide GPS+, not GPS. The longer protection is regulatory: 12 years of US biologic exclusivity if GPS is approved, which would run to ~2039-40. We hold peak sales until then. In the simulation, the exclusivity date barely moves the value.
What the market is paying for
Comparables make the gap concrete. Exhibit 8 sets SELLAS against listed AML companies and two large AML-centered deals from 2020. Syndax already sells an approved menin inhibitor, Revuforj, which brought in $60.7M in 3Q26, plus a second approved drug. Its enterprise value in mid-October was ~$1.19B, about a third of SELLAS's ~$3.54B. Kura, also with an approved menin inhibitor, had an enterprise value of ~$0.36B. Only Gilead's $4.9B purchase of Forty Seven sits above SELLAS, and that's a cautionary tale: it was paid on early data for a drug Gilead later dropped.
SELLAS's EV is ~3x Syndax's, which already sells two approved drugs
Enterprise value for listed companies, headline deal value for acquisitions. Before approval, SELLAS sits between two large AML-centered deals from 2020.
Syndax and Kura EVs are a mid-October 2026 snapshot and move daily.
The closest analog is a vaccine that won on survival and lost commercially. Dendreon's Provenge lowered the hazard of death by 22% in its Ph3 (HR 0.78, median 25.8 vs 21.7 months), in a much bigger prostate cancer market. Logistics, price and a modest absolute benefit capped uptake, and Dendreon was bankrupt within five years of approval. GPS is simpler to give, as an off-the-shelf injection, and its relative benefit is bigger. But it reaches far fewer patients, and the lesson stands: an OS label alone doesn't make a franchise. GPS is also the first WT1 vaccine to win a Ph3. The last one to get there, Sumitomo's DSP-7888, was stopped for futility in glioblastoma in 2021.
Where we stand, and what would change it
We're cautious at $17.95, but this isn't a short. Short interest was already ~28% before the readout, two SLS009 readouts land within weeks, and a buyer could bring its own, more generous inputs. Our self-launch value of $5.48 is ~69% below the close, and our acquirer value of $9.20 is ~49% below it; against the $13.00 deal price the gaps are ~58% and ~29%. Exhibit 9 sums up our confidence: high in the result, low in the price. The events that would close the gap in either direction are the full data, the offering overhang and the FDA meeting, set out below.
High confidence in the result, low confidence in the price
Where we stand on the five questions that decide SELLAS's value from here.
| Question | Confidence | Why |
|---|---|---|
| REGAL met its survival endpoint | High | Randomized trial, OS endpoint, p=0.035, 9.3-month gain in medians |
| The result holds up in the full data | Medium | CI undisclosed (simulated ~0.36-0.97); open-label; three-quarters of listed sites outside the US |
| A takeout near today's price | Medium | Management will "evaluate every path"; on our inputs even $13.00 needs most eligible patients treated, but a buyer brings its own |
| SLS009 frontline succeeds | Medium-low | Encouraging single-arm data; preliminary frontline Ph2 data due Q4 |
| Our base case supports $17.95 | Low | $5.48-9.20 a share; under 1% of simulated acquirer cases clear $17.95, ~10% clear $13.00 |
Bottom line: REGAL met its endpoint, but $17.95 needs more patients, share and price than our funnel and simulations support. On the same basis as our value, even the $13.00 deal price needs most eligible patients treated.
What would move us toward the price.
- Full REGAL data in all randomized patients that confirm a hazard ratio near 0.59, with curves that stay apart and no sign of early harm.
- Evidence that remissions on GPS last well beyond two years, which lengthens the course.
- Pricing signals at $400k or more for a first year of treatment.
- A frontline SLS009 result that supports a registrational path: complete remissions (with or without full count recovery) clearly above the venetoclax-azacitidine control arm in each randomized cohort, without more early deaths.
- A disclosed bid, which would change the question altogether.
What would move us further away.
- The reported analysis turns out to be a subset rather than all randomized patients.
- Transplant or subsequent-therapy imbalances emerge.
- Results or the BAT mix differ sharply by region. 25 of the 73 listed sites are in Greece, Poland, Hungary and Serbia, so outcomes and control treatment by region matter.
- FDA wants more than one open-label trial.
The calendar. Next come the offering's close (Nov 20), SLS009 relapsed-AML data at ASH (Dec 12-15) and preliminary frontline SLS009 data (Q4; cohorts and follow-up undisclosed). After that, an FDA meeting on what SELLAS calls an accelerated filing strategy, and a BLA targeted for 1H27. A full REGAL presentation at a medical meeting matters most for the evidence.
Appendix
Three supporting exhibits: how the company's own patient count compares with ours, how crowded the short side was going into the readout, and how the stock traded around it.
SELLAS's 10-K sizes ~3,750 US patients a year; we get ~960
The 10-K's 8,750 across the US, EU4 and Japan counts half of all second-line patients as reaching a second remission and skips the transplant filter REGAL applied at entry. Neither paper it cites reports a 50% rate.
SELLAS short interest rose more than tenfold into the readout
Shares sold short reached 57.1M on 30 October 2026, the last settlement date published before REGAL topline: ~28% of basic shares outstanding and ~7 days of average volume to cover.
The stock and the short side ran together. SELLAS bottomed at $0.51 in January 2024 and ran to $15.46 by 21 August 2026, ~4.7x in the year before the readout against ~27% for XBI. It then lost about half its value into REGAL while short interest stayed near its highs, 54-63M shares through the summer and autumn. The 17 November jump took the stock to ~38x its 2024 low. A crowded short going into a binary readout may have amplified the move, which is one more reason not to read the whole jump as a verdict on value.
SELLAS ran to $15.46 in August, halved into REGAL, then jumped 156%
Indexed to 100 on 6 November 2023, the start of the short-interest series. In the year before the readout SELLAS rose ~4.7x against ~27% for XBI; it ended at ~1,850 on the index, XBI at ~210.
Sources
- SELLAS, REGAL topline release, 17 Nov 2026 (8-K Ex. 99.1)
- SELLAS, pricing of $150.0M underwritten public offering, 18 Nov 2026
- SELLAS, preliminary prospectus supplement (424B5), 17 Nov 2026
- SELLAS, FY2025 Form 10-K
- SELLAS, REGAL design update, 2022 (8-K Ex. 99.1)
- SELLAS, investor day presentation, November 2019 (hosted copy)
- SELLAS, REGAL update: 72 events, 29 Dec 2025
- SELLAS, corporate presentation, February 2026
- SELLAS, SLS009 ASH 2026 abstract release
- SELLAS, first patient in frontline SLS009 trial
- Jamy O, Cicic D. REGAL protocol. Future Oncol 2024;21:73-81
- ClinicalTrials.gov, NCT04229979 (REGAL)
- Brayer J et al. WT1 vaccination in AML (Moffitt pilot). Am J Hematol 2015
- Wei AH et al. Oral azacitidine maintenance (QUAZAR AML-001). N Engl J Med 2020
- American Cancer Society, AML key statistics
- Mangan JK, Luger SM. Salvage therapy for relapsed or refractory AML (summarizing Armistead et al. 2009). Ther Adv Hematol 2011
- Ferrara F et al. Relapsed AML in the elderly. Haematologica 2001
- Stahl M et al. Azacitidine plus venetoclax in relapsed or refractory AML. Blood Adv 2021
- Angotzi F et al. Venetoclax plus HMA in relapsed or refractory AML (Padua). Front Oncol 2024
- Kurosawa S et al. AML in second remission without transplant. Haematologica 2010
- DiNardo CD et al. Ivosidenib in IDH1-mutated relapsed or refractory AML. N Engl J Med 2018
- Verma D et al. Late relapses in AML. Leuk Lymphoma 2010
- Bottomly D et al. Beat AML 2.0. Cancer Cell 2022 (clinical workbook, CC BY 4.0)
- MD Anderson series, outcomes after venetoclax failure. Haematologica
- Fatima et al. Outcomes after frontline venetoclax-HMA failure (Mayo Clinic). Blood Cancer J 2026;16:138
- Bristol Myers Squibb, FY2023 Form 10-K (Onureg sales)
- OncLive, GPS plus nivolumab in mesothelioma
- Syndax, third-quarter 2026 results
- Fierce Biotech, Gilead to acquire Forty Seven for $4.9B
- Kantoff PW et al. Sipuleucel-T (IMPACT). N Engl J Med 2010
- SELLAS, third-quarter 2026 results and corporate update, 12 Nov 2026 (8-K Ex. 99.1)
- ClinicalTrials.gov, NCT04588922 (SLS009)
- Daver N et al. ENHANCE-3: venetoclax and azacitidine plus magrolimab. Blood 2025
- Sumitomo Pharma, WIZARD 201G (DSP-7888) termination, 15 Dec 2021
- FINRA, equity short interest data
- Yahoo Finance, SLS and XBI daily price history
Modeled figures come from the Clinaptis Analytical Model and Clinaptis Simulation Model and are labeled as such. They are not company guidance.