How a list of companies that would want a new molecule is read out of the clinical trial register, one decided fact at a time, with one company followed from its first trial to its score.
Sources: ClinicalTrials.gov, FDA Orange Book and Purple Book, Drugs@FDA, European and national approval registers, ChEMBL 37
§ 1
The funnel: from the register to six ends.
Most trials never reach a target. Every row that does ends somewhere.
The trial register lists 604,331 studies. Of these, 71,770 were run by a company and can be linked to a drug. Together with the approval registers, those trials give 17,976 drugs with a record, and 5,400 of them name a specific target. Setting each drug’s company against each target gives 12,650 company-target rows, one company at one target.
Figure 1. From the register to the six ends
Registered clinical trials604,331
Company-led trials linked to a drug71,770
Drugs with a record17,976
Drugs naming a specific target5,400
Company-target rows12,650
4,815 company-level needs
Route A2,799
Failed or went quiet; still in the field.
Route B2,016
Marketed drug, no follow-on of its own.
Engaged3,121
The company’s drug here is still active.
Undetermined2,746
A deciding fact is not yet established.
Folded1,674
Many-target drug, counted once.
Excluded294
One sourced fact from a fixed list.
Fig. 1. Bars in the upper part are on a logarithmic scale between 1,000 and 604,331; the counts are exact. The lower bar splits the 12,650 company-target rows into the six ends, to scale. Route A (2,799) and route B (2,016) together are the 4,815 needs. Edition of 20 October 2026.
Every row ends in exactly one of six ways. Two of them are needs, the two routes of the next section. Three say why a row is not one: the company’s drug at the target is still active (3,121), a fact that would decide the route is not yet established (2,746), or the drug acts on many targets at once and is counted once, in its own row (1,674). Only 294 rows are removed, and each is removed by one sourced fact from a short, fixed list: for example a company that dissolved with no successor.
§ 2
The two routes, each with one example.
A company reaches the list by one of two routes.
On route A, the company ran a drug at this target, it failed or went quiet, and the company is still in the field. On route B, the company has a marketed drug here and no successful follow-on of its own. Both say the same thing from different sides: here is a company with a documented reason to want something new at this target.
Route A2,799 rows on 858 targets
Merck & Co. at CTSKcathepsin K, third of 2,799 route A rows
DrugOdanacatib, phase 3.TrialNCT00529373, 2007 to 2017, 16,071 participants. Ended early; the registry records: “The study was terminated early due to an observed increase in risk of stroke”. The main endpoint was met.NowNo active drug at the target; Merck & Co. is still in the field.
Score0.741
Biology proven in humans0.80
Company care0.76
Room to act0.67
Route B2,016 rows on 522 targets
Novartis at FDPSfarnesyl pyrophosphate synthase
DrugPamidronic acid, approved in 1991 and still marketed.Follow-onZoledronic acid anhydrous came later at the same target. Whether Novartis originated it is not yet established, so the follow-on is marked unclear and the row stays a route B need.SourceApproval register entry, 1991; read 20 Oct 2026.
Score0.817
Biology proven in humans0.88
Company care0.92
Room to act0.67
Fig. 2. One row from each route. The three bars are the three questions a score weighs: is the biology proven in humans, how much does this company care, is there room and a reason to act now. Sources: ClinicalTrials.gov, FDA Orange Book, Purple Book and Drugs@FDA, ChEMBL 37; edition of 20 October 2026.
§ 3
Route B rows with a date on them.
When the clock is running
A route B company has a marketed drug and no successor, and for many of them that drug is protected in the United States only for a while longer. Of the 2,016 route B rows, 768 have US protection still running. 118 of them lose it within two years and 256 within four. For these companies the need for a new molecule has a date.
Figure 3. US protection ends, next four yearsroute B · nine of the 256
Company at target · drug
2027202820292030
Ends
Boehringer IngelheimDPP4Linagliptin0.722
20 Dec 2026
TakedaPTH1RParathyroid hormone0.786
23 Jan 2027
NovartisMMESacubitril0.729
27 May 2027
NovartisIL1BCanakinumab0.720
16 Jun 2027
PfizerF10Apixaban0.749
17 Oct 2028
AmgenPSMB5Carfilzomib0.785
22 Nov 2028
Merck & Co.HCRTR1Suvorexant0.743
20 Nov 2029
NovartisCALCRLErenumab0.753
17 May 2030
AstraZenecaEEF2Moxetumomab pasudotox0.723
13 Sep 2030
Fig. 3. Route B rows placed by the date their US protection ends, from the edition date (20 October 2026) to four years on: the nine highest-scoring of the 256. The number after each drug is the row’s score. The dashed line marks two years. A hollow dot is a computed date. US protection ends = the latest counted date: the last listed US drug-substance patent or exclusivity in the FDA Orange Book; for a biologic, its reference-product or orphan exclusivity in the Purple Book; where neither is listed, 12 years from the first US licence, shown as computed.
A protection date is not the same as the date generics or biosimilars can enter.
§ 4
The decision trail for one row.
A rank never stands alone. Here is the trail behind Merck & Co. at CTSK.
The row is traced the way every row is: the route, the facts that decided it with their sources and dates, whether any removing fact applies, and what each part of the score contributed.
Figure 4. Merck & Co. at CTSKRoute Ascore 0.741
RouteMerck & Co. stopped odanacatib at cathepsin K. Its phase 3 trial NCT00529373 stopped on 1 February 2017, and it has no active drug at this target.Decided by
Answered
The failure rests on a stopped trial, a market withdrawal or the drug going quietClinicalTrials.gov, NCT00529373, stopped 1 Feb 2017; read 20 Oct 2026
Answered
The drug has no approval that is still on the marketApproval and market-withdrawal registers; read 20 Oct 2026
Answered
The failure stands: the company's own trial stopped, and another company holds the approval, or the molecule is not an approved drugChEMBL 37 highest phase and the approval registers; read 20 Oct 2026
ExclusionNone. No fact from the closed list of removing facts applies, so the row is ranked.What moved the score
Proof in humans at the target+0.206
Few other companies active at the target+0.172
Active at the target or its disease area recently+0.136
Took its own drug far in human testing+0.125
Trials at the target more often meet than miss their main endpoint+0.092
Why its own drug stopped (not established, neutral)0.000
Fig. 4. The decision trail for one row. Each bar is how much one factor added to the score; 0.000 means the fact is not established and counted as neutral. Sources: ClinicalTrials.gov, the approval and market-withdrawal registers, ChEMBL 37; read for the edition of 20 October 2026.
§ 5
What is not yet known.
The list says where it does not know, and does not guess.
An unknown is a value, not a gap. It counts as neutral, it is named on the row, and it can change when the fact is established. There are three kinds.
In one rowWhy Merck & Co. ended its work on the drug as a wholeThe registry gives a stated reason for one trial, but the reason for ending the whole drug is not on record. The factor counts as neutral (0.000 in Figure 4), and the row says so.
In a routeWho originated the later drug at FDPSZoledronic acid anhydrous came after pamidronic acid, and whether Novartis originated it is not established, so the follow-on is marked unclear. 2,746 rows carry such an open fact, each naming it.
At the edge of the dataWhat the registers do not holdSome approved drugs have no approval record in the registers read, and some subsidiary names are not yet placed in their parent group. Drugs that no public register lists are not seen.
Fig. 5. Three kinds of unknown, with a case of each from this paper.
§ 6
How an agent uses this.
An agent reads the same trail, and cites it.
Ask an agent the question the list was built for and it works over the facts above rather than from memory. It resolves the target, finds each company that ran a drug there, applies the routes, and writes an answer in which every line points at the fact it rests on. It recommends; it does not decide. The route, the exclusions and the rank are fixed by the facts, not by the agent.
Figure 6. An agent task, answereddone · 4 steps
TaskWhich companies would want a new molecule for CTSK?
What the agent did
1Resolved CTSK to the protein cathepsin K and listed every drug tested at it.
2Found each company that ran one, and what became of its drug.
3Kept the companies still in the field with no active drug here, after checking the closed list of removing facts.
4Ranked them on three questions and attached the deciding fact to each line.
Answer, first of five
Cited
Merck & Co. ranks first for CTSK at 0.741, on route A.Figure 4 · score and route
Cited
Its phase 3 trial of odanacatib stopped on 1 February 2017.NCT00529373 · stopped 2017
Cited
Odanacatib has no approval still on the market, and Merck & Co. has no active drug at the target.Approval registers · 20 Oct 2026
Open
Why Merck & Co. ended its work on the drug as a whole is not yet established; the score treats it as neutral.Open · stop cause
Fig. 6. The task from the product page, answered for the first company only. The other four are Novartis (0.612), GSK (0.549), Medivir (0.547) and Ono Pharmaceutical (0.521), all route A. Each chip names the fact behind the line.
§ 7
Method and limits.
How the list is built, how its rows are ranked, and what it does not yet know.
What the list is, what removes a row, what “undetermined” means, how the three questions are weighed, which sources it reads and what it is known to miss are set out on the method page. Every number on this page is from the edition of 20 October 2026.
Bring a target or a company. We will run it with you.
In 30 minutes we give an agent the target, company or area you are looking at and read the answer together, with the trail behind every line. We reply within one working day.