Six of 85 proteins are both selectively required and drugged — and MEN1 is not close to the others
also called two-axis target triage · selectivity x ligandability
We already knew, separately, which proteins this leukaemia needs more than other leukaemias do, and which proteins anyone has ever made a molecule for. Putting the two lists side by side sorts every protein in the graph into four groups. Only six are in the group you could start work on tomorrow, and fifteen are things the disease clearly needs that nobody has a molecule for at all.
The same thing, for a scientist
Joining the persisted 17,728-gene selectivity scan to the ligandability census gives 85 proteins with both measures. Taking selective as rank ≤500 and ligandable as ≥5 distinct co-crystallised drug-like ligands: **6 are both** (MEN1, FLT3, BCL2, KAT7, DOT1L, MLLT1), **15 are selectively required with no chemical matter**, 16 are drugged without ranking selectively, and 48 are neither. MEN1 is rank 1 with 45 ligands and is not close to anything else on either axis.
Why cross them, and the calibration that says how far to trust it
Each axis alone is a weak filter. Selective requirement says the disease needs a protein without saying anyone can touch it; ligandability says a protein can be touched without saying it matters here. Crossed, they separate four situations that call for different work.
The calibration is FLT3, and it is a warning about the second axis. FLT3 ranks 2 of 17,728 for selective requirement and has 5 co-crystallised drug-like ligands — while three FLT3 inhibitors are approved and in clinical use. A clinically drugged kinase scoring 5 shows the ligand count is a floor, not a measure: approved drugs do not always leave deposited co-structures. Read a low count as no public structural evidence of chemical matter, never as undruggable.
Both axes: six proteins
| gene | selectivity rank | drug-like ligands | % construct covered |
|---|---|---|---|
| MEN1 | 1 | 45 | 100.0 |
| FLT3 | 2 | 5 | 94.3 |
| BCL2 | 36 | 31 | 91.2 |
| KAT7 | 92 | 5 | 45.2 |
| DOT1L | 184 | 23 | 37.3 |
| MLLT1 | 487 | 12 | 26.5 |
MEN1 is not close to the others. Rank 1 with 45 distinct ligands; the next best combination is BCL2 at rank 36 with 31, and BCL2 is an apoptotic node rather than anything fusion-specific. This is the third independent line in this graph arriving at the same place — the interface argument from Yokoyama 2004, the ligandability census, and now this — and all three are ultimately reading the same small literature, which is the reason to keep saying it rather than treating it as settled.
Selectively required, no chemical matter: the honest opportunity list
15 proteins: AFDN, AFF1, CBFB, CCND3, ELL, GFI1, LEO1, MEF2C, MEIS1, MLLT10, NSD1, PSIP1, USP22, ZEB2, ZFP64.
| gene | rank | ligands | % covered |
|---|---|---|---|
| PSIP1 | 7 | 4 | 100.0 |
| MEF2C | 8 | 0 | 0.0 |
| ZEB2 | 20 | 0 | 4.9 |
| ZFP64 | 32 | 0 | 21.2 |
| GFI1 | 78 | 0 | 0.0 |
| MEIS1 | 84 | 0 | 14.9 |
| AFF1 | 160 | 0 | 3.5 |
| USP22 | 226 | 0 | 0.0 |
Two of these are the project's own load-bearing proteins.
AFF1 is the fusion partner itself — rank 160, zero ligands, 3.5% of the chain ever present in a deposited construct. The molecule that drives the disease is both structurally dark and chemically untouched.
PSIP1/LEDGF is rank 7 with 4 ligands, just under the threshold, and fully covered structurally. It is the tether in the tether-and-pocket model, it is the 7th most selectively required gene in the genome for this disease, and it has almost no chemistry — while the graph separately records that no published experiment removes LEDGF from an established human KMT2A::AFF1 leukaemia. High selectivity, near-zero chemical matter, and the key experiment unrun.
The rest are transcription factors and chromatin proteins — MEF2C, ZEB2, ZFP64, GFI1, MEIS1 — where zero ligands is unsurprising and the realistic modality is degradation or interface disruption rather than orthosteric inhibition.
Drugged without a rationale here
16 proteins: ABL1, ALK, BRD4, CCNT1, CDK9, CREBBP, CTNNB1, HCFC1, KAT8, KDM5A, KRAS, NRAS, PTPN11, RARA, SIRT1, WDR5.
These have real chemistry and do not rank as selectively required in KMT2A-rearranged lines. CDK9, BRD4, SIRT1, KRAS and ABL1 are all heavily prosecuted and all sit outside the top 500 on the selectivity axis.
This quadrant is the standing argument against the drug what is druggable reflex, and it is where a project under time pressure drifts. It is also the quadrant where this analysis is weakest: a target can be worth combining without being selectively required alone, which is exactly the objection recorded against the countermeasure ranking.
| script | analyses/target-triage-two-axes/run.py |
|---|---|
| status | executed |
| run date | 2026-09-19 |
| n joined | 85 |
| headline | 6 of 85 are both selectively required and drugged. MEN1 (rank 1, 45 ligands) is not close to the others. 15 are selectively required with no chemical matter, including AFF1 and PSIP1/LEDGF. |
| caveat | The ligand axis is a FLOOR, not a measure - FLT3 ranks 2nd for selectivity, has three approved inhibitors, and scores five co-crystallised ligands. Absence of deposited co-structures is not absence of chemistry. The selectivity axis has its own limits: eleven KMT2A-rearranged lines, none infant-derived, CRISPR knockout rather than pharmacological inhibition, and a difference statistic that is least informative for strongly pan-essential genes. Both thresholds are arbitrary lines through continuous quantities and the quadrant labels should not be read as categories. |
thresholds
{
"selective_rank": 500,
"ligandable_min_ligands": 5
}quadrant counts — the numbers behind the chart
{
"selective_and_ligandable": 6,
"selective_no_chemistry": 15,
"ligandable_no_rationale": 16,
"neither": 48
}quadrants
{
"selective_and_ligandable": [
"BCL2",
"DOT1L",
"FLT3",
"KAT7",
"MEN1",
"MLLT1"
],
"selective_no_chemistry": [
"AFDN",
"AFF1",
"CBFB",
"CCND3",
"ELL",
"GFI1",
"LEO1",
"MEF2C",
"MEIS1",
"MLLT10",
"NSD1",
"PSIP1",
"USP22",
"ZEB2",
"ZFP64"
],
"ligandable_no_rationale": [
"ABL1",
"ALK",
"BRD4",
"CCNT1",
"CDK9",
"CREBBP",
"CTNNB1",
"HCFC1",
"KAT8",
"KDM5A",
"KRAS",
"NRAS",
"PTPN11",
"RARA",
"SIRT1",
"WDR5"
],
"neither": [
"AFF2",
"AFF3",
"AFF4",
"ASH2L",
"BCOR",
"BCR",
"BOD1L1",
"CBFA2T3",
"CBX8",
"CDC73",
"CTR9",
"DPY30",
"ELL2",
"ELL3",
"EPS15",
"ETV6",
"EWSR1",
"FLI1",
"GLIS2",
"HLF",
"HOXA9",
"ICE1",
"KDM2B",
"KDM4C",
"KDM6A",
"MED26",
"MLLT3",
"MLLT6",
"MYH11",
"NF1",
"NTRK3",
"NUP98",
"PAF1",
"PBX1",
"PML",
"PPIE",
"PRMT1",
"RBBP5",
"RTF1",
"RUNX1",
"RUNX1T1",
"SETD1A",
"SGF29",
"SIAH1",
"SIAH2",
"SKIC8",
"TCF3",
"TRRAP"
]
}calibration flt3
{
"rank": 2,
"ligands": 5,
"note": "three approved inhibitors; the ligand axis is a floor, not a measure"
}- A two-axis grid looks like a decision procedure and is not one. Both thresholds are arbitrary, PSIP1 misses 'ligandable' by a single ligand, and presenting quadrants invites exactly the categorical reading the caveats forbid.
- Reporting that MEN1 stands alone for the third time from a third method risks manufacturing confidence. The three lines are not independent - they read the same small literature and the same cell-line panel - and the appearance of triangulation is itself a bias worth naming.
- The quadrant labelled 'drugged without a rationale' contains CDK9, BRD4 and SIRT1, all with active programmes. Using this table to argue against those programmes would be over-reading a drug-naive single-gene screen.
- PSIP1/LEDGF is rank 7 with almost no chemistry and the decisive experiment unrun. Is the absence of ligands a record of failure or of nobody trying? A patent and screening-deck search separates those and they imply opposite decisions.
- Would the quadrants move if the selectivity scan were run in lines made resistant to a menin inhibitor rather than drug-naive ones? Every acquired dependency is invisible here.
- Only 85 of 565 entries carry both measures. What is in the other 480 - and is the missing measure selectivity, ligandability, or a UniProt accession nobody recorded?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Across 85 proteins with both measures, 6 are selectively required (rank ≤500) AND carry ≥5 co-crystallised drug-like ligands: MEN1 (rank 1, 45 ligands), FLT3 (2, 5), BCL2 (36, 31), KAT7 (92, 5), DOT1L (184, 23), MLLT1 (487, 12). 15 are selectively required with no chemical matter, including AFF1 (rank 160, 0 ligands, 3.5% covered) and PSIP1/LEDGF (rank 7, 4 ligands). 16 are drugged without ranking selectively.
Two-axis target triage: KMT2A-rearranged selectivity crossed with demonstrated ligandability · Oncofusion knowledge graph; DepMap via Open Targets + RCSB PDB 2026What this evidence cannot show
Computed by the coordinator on 2026-09-19 by joining two already-persisted analyses on gene symbol; no new measurement. BOTH THRESHOLDS ARE ARBITRARY LINES THROUGH CONTINUOUS QUANTITIES - PSIP1 at 4 ligands falls outside 'ligandable' by one - and the quadrant labels should not be read as categories. The ligand axis is a floor (see FLT3). The selectivity axis rests on eleven KMT2A-rearranged lines, none infant-derived, uses CRISPR knockout rather than pharmacological inhibition, and is least informative for strongly pan-essential genes. Only 85 of 565 entries carry both measures, so this is a triage of what happens to be measured.
Evidence that argues against this (1)
- The framework rewards single-agent lethality and would therefore rank a good COMBINATION partner near the bottom - a protein whose loss does nothing alone and everything alongside menin inhibition scores zero on the axis that matters most for how these drugs are actually used. The same objection sank half the countermeasure ranking. It also cannot see conditional or acquired dependencies: anything that only matters after menin-inhibitor exposure is invisible in drug-naive lines by construction. Oncofusion knowledge graph; DepMap via Open Targets + RCSB PDB 2026
Same comparator; the selectivity axis is conservative.
One of the two axes joined here.
- related to →The whole protein-coding genome: menin is still first, and a metabolic pathway appears (run)
The other axis; supplies per-gene ranks.
- related to →Six countermeasures, no head-to-head — so rank their targets: CDK6 is 13th and CDK4 is 16,052nd
Same selectivity axis, applied to proposed combination partners - and carrying the same objection, that single-agent lethality is the wrong test for a combination.
Rank 1 with 45 distinct drug-like ligands; the only protein strong on both axes and not close to the others.
Rank 160 with ZERO ligands and 3.5% of the chain ever in a deposited construct - the molecule driving the disease is structurally dark and chemically untouched.
PSIP1 is the 7th most selectively required gene in the genome for this disease, fully covered structurally, with 4 ligands - and the experiment removing LEDGF from established disease is unrun.
Supplies the druggability triage the phase-2 dossier needs, on two axes and with the FLT3 calibration stated.