Michael Pace

@deathtouch2k Great stuff. 450/150 PY your est exposure adj for 50mg/25mg in pt 2? Headline imbalance to be expected at that rate then, w exposure adj more inline?

Based on our conversations with the buy side and sell side, one thing is clear: the market is still struggling to define the actual safety bar for $ABVX. Specifically: how many non-NMSC events would ABTECT Maintenance Part 2 need to show before FDA reviewers or sophisticated pharma BD teams should become concerned that obefazimod may have a true carcinogenic signal, rather than background malignancy events in a refractory UC population? We provide our analysis and conclusions below. Said simply: under our assumptions, $ABVX would need to show a malignancy burden far beyond one or two incremental cancer cases before the bear case becomes statistically and biologically coherent. When a drug trial reports a handful of cancer cases, the natural question is: is this a real signal, or just bad luck? A drug study with 150 patients over one year will almost certainly see some cancers — because UC patients get cancer at a background rate even without any drug. In fact, even otherwise healthy people get cancer at a background rate without any drug. The real question is whether the number of cancers seen on the drug is more than we would expect by chance from that background rate alone. Enter statistics: the formal language of separating signal from bad luck. In this case: what is the event burden and pattern that is high enough, coherent enough, and biologically plausible enough to overcome a low causal prior to obefazimod being carcinogenic? This is exactly the kind of question Bayesian inference is useful for answering. Our Bayesian inference model asks a simple question: Given an estimated background non-NMSC incidence rate of ~0.5/100 PY, how many additional non-NMSC events would need to appear in the next ~450 PY of 50 mg exposure before the pooled 50 mg maintenance dataset begins to look meaningfully above UC background? At ~0.5/100 PY, the expected background incidence over ~450 PY is roughly 2–3 non-NMSC events. Our model suggests the upcoming 50 mg Part 2 update would need to show something closer to 10 additional events above background — roughly 14 total non-NMSC events in the pooled ~600 PY 50 mg maintenance dataset — before the posterior begins to suggest a >50% probability that the true obefazimod non-NMSC rate exceeds the upper bound of UC background. We favor a Bayesian inference framework in these cases because it contemplates the data through a lens similar to that of what we believe an FDA reviewer and pharma regulatory teams would look through: exposure-adjusted rates, confidence intervals around those rates, event heterogeneity, organ clustering, baseline UC cancer risk, prior therapy risk, timing of exposure, investigator attribution, and whether the pattern resembles a broader immunosuppressive AE phenotype. Most importantly, with this approach we do not need to assume obefazimod causes cancer. In fact, the causal prior should be low because there is no obvious a priori hypothesis for obefazimod causing cancer, including clean preclinical work, no clustering to a specific cancer type, no broader adverse event pattern associated with impaired immune surveillance, and investigator assessment that the disclosed prostate and breast cancer cases were unlikely related. Thus, the analysis is not whether “observed cancers divided by patient-years is greater than zero.”; that is coarser than the data allows. The question is whether the observed event pattern is adequate to overcome a low causal prior. We estimated background malignancy IRs across an aggregate 7000+ PY based on a synthesis of Colombel et al. (2017), Rubin et al. (2026), D’Haens et al. (2023), Swissmedic (2024), Sands et al. 2026, Abreu et al. (2022), and Sandborn et al. (2019), and arrived at an estimated non-NMSC IR of 0.5, which is the midpoint of the range provided by management in the Part 2 primer deck published earlier today (ir.abivax.com/static-files/7…). We use 0.5/100 PY as the non-NMSC IR as the background prior of our model. Our analysis allows us to use 7000 PY as the weight of that prior probability. Then, we ask how many non-NMSC events need to be observed in the ~450 PY from Part 2 for the aggregate obefazimod non-NMSC IR to indicate that the true obefazimod non-NMSC rate exceeds the UC patient background upper bound with at least 50% probability. We set the PY reported in ABTECT Maintenance Part 1 according to management feedback. An estimated background non-NMSC IR prior of 0.5/100 PY is at the midpoint of the range provided by $ABVX management and implies an expected 2-3 non-NMSC events per 450 PY. Our model suggests that the ABTECT Maintenance Part 2 update for the 50 mg dose would need to show 10 additional events over and above those background events in 450 PY (i.e., total of 14 non-NMSC events) before the pooled 600 PY maintenance dataset (Part 1 + Part 2) for the 50 mg dose to indicate a greater than 50% probability that the true obefazimod non-NMSC rate is greater than the upper bound of UC background rate, when the observed non-NMSC IR begins to possibly be considered elevated above background expectations. Even when we cut the prior probability weight in half to 3500 PY, the conclusions remain unchanged: ABTECT Maintenance Part 2 update for the 50 mg dose would need to show 7 additional non-NMSC events over and above the number of events expected given the background IR before the pooled 600 PY maintenance dataset (Part 1 + Part 2) for the 50 mg dose to indicate a greater than 50% probability that the true obefazimod non-NMSC rate is greater than the upper bound of UC background rate. Said differently, the key risk is not that another cancer event occurs; the UC and IBD literature documents an expected incident rate for non-NMSC at their background rate (0.3-0.7/100 PY). The key question is whether the next ~450 PY expected from the Part 2 50 mg arm is consistent with the drug-driven malignancy hypothesis: repeated non-NMSC events, organ clustering, biological patterning, exposure-duration logic, or a broader immune-surveillance signal. That is what our model tests. If the next cut instead shows low-count, disparate, background-plausible events, the current obefazimod-driven malignancy argument becomes increasingly difficult to sustain. But what about Rinvoq? They didn’t see a big imbalance? This is not a Rinvoq/Xeljanz-style RA safety signal, where a nearly 4,500-patient outcomes study produced a statistically significant malignancy and immunosuppressive AE cluster in line with an a priori safety hypothesis prior. So far, these few malignancy observations are easily explainable by the background rate of UC cancer in this population especially, and mechanistically, there is also no obvious reason to believe obefazimod should be carcinogenic. If anything, Qin et al. (miRNA-124 in Immune System and Immune Disorders - PMC) discuss miR-124 biology in anti-proliferative and anti-fibrotic contexts, though we would not underwrite an anti-cancer effect here. Against the magnitude of efficacy demonstrated by obefazimod to date: ~50% clinical remission versus ~10% placebo, plus large effects on endoscopic remission, HEMI, steroid-free remission, and sustained remission that is life-altering; these non-NMSC events do not appear to alter the risk-benefit equation. We do not think these disclosed data support a black box. A black box for what appears likely to be background artifact would risk causing net harm to patients by creating unnecessary friction around delivery of a novel and highly efficacious therapy. A brief comment on NMSCs: Based on our diligence, including conversations with former FDA review leadership responsible for evaluating new IBD drug approvals, NMSCs are most commonly basal cell carcinoma and cutaneous squamous cell carcinoma. Importantly, these cancers are: 1) very common; 2) very slow growing; 3) very visible; and as a result, 4) very treatable and curable. These cancers are rarely metastatic, and very rarely fatal. There is published evidence that IBD patients appear to be at elevated rates of NMSCs, hypothesized to result from either higher baseline persistent inflammatory processes and/or thiopurine use. In fact, clinical guidelines published in April 2025 following an expert panel from the American Gastroenterological Association Clinical Practice Update provided specific guidance that all adult patients with IBD should follow skin cancer primary prevention practices by avoiding excessive UV exposure, that patients on immunomodulators, anti-TNF biologic agents, or small molecules should undergo yearly total-body skin exam, and that patients with any history of thiopurine use should continue yearly total-body skin exam even after thiopurine cessation. In the cohort study, the incidence of NMSC was higher among patients with IBD compared to controls (IRR 1.64, 95% CI 1.51–1.78). Recent thiopurine use was associated with NMSC (adjusted OR 3.56, 95% CI 2.81–4.50), and persistent thiopurine use was associated with NMSC (adjusted OR 4.27, 95% CI 3.08–5.92). While, prior to our analysis, we were not concerned that the FDA or pharma would view elevated NMSC rates as a major issue, we were concerned that investors might continue to misclassify them as a serious systemic malignancy signal. However, based on recent conversations with other buysiders, we sense the market is beginning to understand that NMSCs are not a meaningful issue here: they are generally detectable, manageable, and already incorporated into the standard-of-care dermatologic monitoring framework physicians use for UC patients. For completeness, we repeat our Bayesian inference analysis as described above but for NMSC. Taking the NMSC background IR of 1/100 PY provided by management today, we expect 4.5 cases in a 450 PY dataset. Our model suggests that the ABTECT Maintenance Part 2 update for the 50 mg dose would need to show 12 to 13 additional events over and above those background events in 450 PY (i.e., total of 21 NMSC events) before the pooled 600 PY maintenance dataset (Part 1 + Part 2) for the 50 mg dose to indicate a greater than 50% probability that the true obefazimod NMSC rate is greater than the upper bound of UC background rate, when the observed NMSC IR begins to possibly be considered elevated above background expectations. In summary: The right conclusion is not that no cancer events occurred, or that no events will occur in the future. They will occur, not only because cancer risk is elevated in UC, but because cancer can occur in anyone, at any time, including otherwise healthy patients. We think the right conclusion is that the events that did occur look explainable by baseline patient risk, common cancer epidemiology, UC biology, and routine screening detection, rather than acting as evidence of an obefazimod-specific oncogenic signal. We think FDA reviewers and pharma regulatory teams are more likely to reach this conclusion than to view the disclosed cases as a coherent obefazimod-specific malignancy signal, making increasingly clear what we believe is a best-in-disease efficacy/risk profile in the lucrative maintenance setting for UC and substantially de-risking the company’s opportunity in Crohn’s. Disclosure: I/we may be long ABVX and may buy, sell, hedge, or otherwise change exposure at any time without notice. Not investment advice or a recommendation. Analysis reflects current views and assumptions based on public disclosures, published literature, and non-confidential conversations, and may change as new data become available. No compensation from ABVX or any third party.

@A_May_MD New maintenance primer today too, hopefully people level set expectations ahead of release re: bckgrd rates and exposures ir.abivax.com/static-files/7…

Finally, the first time I’ve seen someone get it right. $ABVX has the BEST IN DISEASE efficacy in maintenance. Not “best in class” (as I’ve seen repeatedly). Obefazimod is the ONLY drug in its class, and its maintenance efficacy is BEST IN DISEASE. Two very important points/distinctions.

@ipollit77 @A_May_MD @shirefindsvalue likelihood of a causal association, it will not be included in labeling.”

@ipollit77 @A_May_MD @shirefindsvalue “Another safety consideration was an imbalance in malignancies in the fezolinetant treatment groups compared to placebo. This finding did not appear to be clinically relevant as many of the cases of malignancy appeared to be pre-existing, there was a short time to onset to the...

$ABVX seems to have (in the last hour or so?) quietly released a new corporate deck with 3 important slides at the end. All, in my opinion, providing strong new information showing that these cancer cases were unrelated to drug and equivalent to expected background noise. Most important is the slide on the 2 non-nonmelanoma skin cancers. BOTH of these were more indolent, low/intermediate subtypes of their respective cancer (prostate and breast). Not only does this mean that the cancers are less threatening, it also means that THEY ARE SLOWER GROWING CANCERS. Why is this particularly important? Because it means that the development of the cancers very (very) likely PREDATES THEIR ENROLLMENT IN THE TRIAL. Look into the doubling times of grade 2 (Gleason 7) prostate cancer and grade 2 NST breast carcinoma. These are slow-growing tumors that very likely existed before these patients ever even had a dose of obefazimod. That relates to another key finding on this slide - the prostate cancer case was identified via PSA screening at 8.5 months into the study (remember earlier is better). In the Guggenheim conference they had said it was confirmed at day 367...they must have been referring to the biopsy confirmation of the subtype, not PSA confirmation of the prostate cancer diagnosis. This new information speaks to an earlier diagnosis. The breast cancer patient was diagnosed even earlier than that! Only 6.8 months of Obe exposure. Also, these new slides give us actual information on the prior drug exposures - before this afternoon we knew that they were on some prior treatments, but we didn't know what...THE PROSTATE CANCER WAS PREVIOUSLY EXPOSED TO ***5*** DRUGS WITH LABELED CANCER WARNINGS BEFORE ENROLLING IN THE STUDY! 3 OF THEM HAD ***BLACK BOX WARNINGS*** FOR CANCER RISK! -Humira (black box) -Infliximab (black box) -Rinvoq (black box) -Entyvio (warnings and precautions) -Stelara (warnings and precautions) We also just got new info on the NMSC cases (which matter far less but which spooked the market anyway). How you can look at the details of these skin cancer cases and think they are related to the drug is beyond me (but then again, these details just got released - quietly, for some reason). First of all, ***ALL OF THE 4 50MG CASES OF NMSC OCCURED IN 6 MONTHS OR LESS!!! Again, too rapid to be reasonably assumed drug-related. The fact that they all happened in the first half of this study is actually extremely exculpating evidence for $ABVX. Other details: -4/5 were 60+ years old (STRONGLY associated with skin cancer risk) -3/5 had PRIOR SKIN CANCER ALREADY(!!!!!) -4/5 had prior exposure to other drugs that are known to increase skin cancer risk. Finally, they also added a slide discussing that some studies have shown the elevated risks of these cancers for UC patients at baseline. -~5x higher risk of prostate cancer in IBD patients -~2x higher risk of breast cancer in IBD patients Why did $ABVX add these 3 slides to the corporate deck randomly, silently, on a Friday afternoon? IDK. Legitimately good news in those slides! I'd have pressed released this info as soon as I had it, because the details really help alleviate the (already statistically misguided) concern that these cancers could've been caused by Obefazimod. Here's the link: ir.abivax.com/static-files/e…

@ipollit77 @A_May_MD @shirefindsvalue development of malignancy, there was a lack of biological plausibility, there were no concerning findings in the nonclinical studies, and there was a variety of cancers without a clear pattern. This imbalance appeared to be a chance finding and given the lack of a reasonable…

@viggy_krishnan @CousinGraig @itsDanielWu In any event as training runs grow into hundreds of billions frontier token costs basically going to limit those models to large enterprises anyways, kind of feels like where we’re heading even absent reg so long as the final outcome not nationalization or outright ban.

@viggy_krishnan @CousinGraig @itsDanielWu If we do go down this path, wouldn’t be shocked if approved US vendors get model access directly via some clearance level and sell products (outputs, applications, models) to broader customers with governance and security around it (req auth, etc). Software bull case?!

@viggy_krishnan @itsDanielWu Chances this is “Art of the Deal” on the equity stakes we just heard about?

How does something like this not torpedo the AI intelligence explosion bull case? USG establishing precedent that access to anything as smart as Fable 5 (which is not RSI and nowhere near AGI) will be banned. Even if Ant could make the model accessible to US nationals, how will any customer ensure compliance seeing as not all employees of US enterprises are US nationals… So we have a situation where the labs need to spend increasing amounts of capex to build more powerful models, but are restricted from monetising them. I do not think the intelligence levels of opus 4.8 and gpt 5.5 are enough to justify anywhere close to the amount of AI capex being spent, let alone projected to be spent…

@ipollit77 @A_May_MD @shirefindsvalue That’s treating biologically diff cancers as a single event without looking at any causal mechanism. What’s the rationale for adding those diff AEs together and nothing else? If you split by diff types and do the same thing, accounting for bckgrd rates, becomes less concerning.

@A_May_MD @shirefindsvalue Yes, based on P3, there is a 1–2% chance that it is pure coincidence. At least that is my op, and so far I haven’t found anything to contradict this simple math. To be clear, it’s not a stat whether obe act poses a cancer risk or might even offer protection.That remains an open q

@Andre_AGTC @DominickCampag Avg across bigger bucket $4-$5bn per indication unadj

@DominickCampag $ABVX projected revenue in UC + CD Leerink 4B usd Stiffel 3B eur JMP 2.6B updated to 2.9B Jefferies 2.6B no update post maintenance Stiffel is the most reliable IMO.

$NUVL aquision at 10B makes the multiple to 4B peak revenue of 2.5 The typical multiple in this price range is 2.3-2.5 It puts $ABVX at 122 BO price assuming 4B peak rev in IBD (UC +CD)

@ContrarianCurse Added to a custom losers basket and trailing momentum flipped negative…their fate is sealed, AI may as well have already replaced them /s

Big cap internet has somewhat same factor as software which I sort of get but also most of those businesses are aggregators / network effect businesses Range of outcomes far more narrow than software and many of them are far cheaper

@A_May_MD @epickram Wasn’t totally clear if “near-term IBD” just UC, but suggested that way. That was pre-P3, so can assume a PoS adj there relative. And “majority”not value not all. But still can argue based on that closer to 4x-5x paid if just going on latest stage indication.

@A_May_MD @epickram I think also it’s hard to know with proxy estimates what mgmt is assigning for PoS vs market and what is each indication (nor granular). In ARNA, PFE said maj of $6.7bn was “near-term IBD” and later “potential launch in UC”. A year later they said peak sales in UC of $1bn-$2bn.

Question on $ABVX valuation for bulls out there (like myself) @A_May_MD @seedy19tron : $MRK and $PFE did their deals at ~1.2x peak risk-adj. sales see merger proxy data: Prometheus: $10.1B EV / $8.162B peak sales = ~1.2x Arena: $6.0B EV / $5.070B peak sales = ~1.2x -cont.

@Merridew__ @quantian1 Prob convert related?

@Lord_of_Biotech @AppleHelix Also quite common for partnership discussions to be a starting point and evolve (if interest is there)

@AppleHelix Thats what La Lettre said, the Kepler notice said partnership or takeover. Everything is on the table for $ABVX

Odd $ABVX trading this morning. Obviously everybody just read the headline from the La Lettre article: It quite drastically speaking about a transaction with big Pharma „after receiving information requests“ which is a due diligence request? Isn’t this article super bullish instead of negative?

@BrokenMoats Term fee $600mm but pays for superior proposal / chg in rec, not a unilateral out on px (specific performance)

@BrokenMoats Nums look right on shares/net debt, although dilution large (BLD proxy pg 186). EBITDA take mgmt proj for QXO (incl Kodiak) - “future unidentified M&A” = $1.4bn. +$1.3bn BLD vs pro forma $2.83bn (so ~$100mm synergies in 27). Floor conditions? Will prorate to mix if that’s Q

$QXO gang anyone have a current cap table, wanted to run some numbers but its hard to keep up with all the issuance/debt offerings.... Pro forma suggests there is around 1.1B shares outstanding? (will that include all the financing to get Topbuild closed (I believe so)? + around $8B in net debt (pro forma) So at $14 a share - that would be roughly $23B EV value? Anyone confirm this or tell me which pieces are wrong (for Kodiak, beacon and Topbuild) Roughly $2B EBITDA off their press release (probably up for debate, rates/economy, then need to look how much synergies might be backed into that) So 4x leveraged (means likely more shares need to get issued to de-risk especially with interest rate risks and transmission to demand) so 11.5x with mote dilution (likely).....getting close but not there yet - still think you need to wait for 10x or < Last and most important question does the math around Beacon change - 15x is meaty for a distributor multiple so unlikely they walk. But with the 45/55 split (is there floor conditions or will TopBuild eat that difference if/when they vote later this month (if not anyone know what the break fee is). Thanks in advance

@BenBajarin Thanks!

@Msp194 Ugh sorry it didn’t include the link. It’s here thediligencestack.com/p/agentic-eda-…

We spent time digging into the value lift Agentic EDA will bring to $CDNS and $SNPS. A story we we think is still misunderstood and under appreciated. Angstrom era + massive giga cycle and many other levers are very good for these two. Full report: thediligencestack.com/p/agentic-eda-…