Literature workflows
Resolve the selective early gene response to neuroactive drugs
Read the task instructions and deliverables
Active neuroactive drugs can reduce glioblastoma growth despite having different clinical targets. Use the supplied genome-wide six-hour expression measurements from LN-229 glioblastoma cells and the sample metadata to identify genes that most clearly support a shared early response to active neuroactive drugs. How distinct is this response from the comparison drug treatments, including temozolomide?
Report the supported gene set as shared_genes. Expression values are the deposited DESeq2 variance-stabilized values, a scale that reduces differences in measurement noise. Use them as supplied, without another transformation. Report shifts in these transformed expression units, not RNA fold changes. For the three numeric fields, use your reported shared_genes: each gene's group shift is the arithmetic mean of that group's samples minus the arithmetic mean of the NEG samples, all at six hours. Use the sample metadata's drug_type labels for NAD, ONCD, and NEG; temozolomide means only samples with drug equal to TEMOZOLOMIDE. Include all supplied replicates with equal weight. Each numeric field is the arithmetic mean of its group shifts across your reported genes, rounded to four decimal places.
In reasoning.md, give 4–8 bullet points explaining the key observations, the inference each supports, and how the comparison groups and possible alternative explanations affect the conclusion. Address variation among drugs and related genes, and what these expression measurements can establish about growth inhibition.
Return EXACTLY, as the contents of answer.json:
{"shared_genes": ["<gene symbol>"], "nad_mean_shift": <float, 4 decimal places>, "oncd_mean_shift": <float, 4 decimal places>, "temozolomide_mean_shift": <float, 4 decimal places>}How was it graded?
Deterministic checks compare the submitted answer fields with the reference values and tolerances. An LLM judge assesses the remaining reasoning criteria and cannot override the deterministic results.
One point per passed criterion, divided by the total number of criteria. Unresolved judgments produce a score range. Critical failures are reported separately and do not add weight. No overall pass threshold is specified.
LLM judge: Not recorded in this packet
Task files
Choose an evidence file.
Grading criteria
Identify genes in the shared early response to neuroactive drugs
Earns credit
Identify the shared early-response gene set for active neuroactive drugs. Submit shared_genes as a JSON list of strings. After alias normalization and removal of optional members, set overlap must be at least 80%. Overlap means shared members divided by all distinct members in either set (Jaccard similarity); both omissions and extra members reduce it. Reference set (19 members): ATF3, BTG1, CXCL8, DNAJB9, EGR1, FOS, FOSB, GEM, HERPUD1, HSPA1A, HSPA5, IER3, IL11, INHBA, JUN, NR4A1, PTGS2, SAT1, TGIF1.
Does not earn credit
No credit for a missing or malformed answer, a field that is not a list of strings, or set overlap below the required threshold. Optional members do not count as errors.
Accepted alternatives
Order does not affect the set comparison. Repeated members are counted once by the scorer. Letter case is ignored. Leading and trailing whitespace is ignored. Accepted aliases: IL8 → CXCL8; IL-8 → CXCL8; c-FOS → FOS; cFOS → FOS; GRP78 → HSPA5; BiP → HSPA5; c-JUN → JUN; cJUN → JUN; COX2 → PTGS2; COX-2 → PTGS2. Optional members, ignored in the comparison: ABL2, AC009597.1, ARC, ASAH1, ATF4, AVPI1, BHLHE40, CD55, CDKN1A, CEBPB, CHAC1, CXCL1, DDIT3, DDIT4, DNAJB1, DUSP1, DUSP6, EPB41L4A-AS1, GADD45B, GARS, GDF15, HBEGF, HSPA1B, HSPA6, HSPH1, IER2, IER5, IFI6, INSIG1, ISG15, ITPRIP, JUNB, JUND, KLF4, KRT7, LDLR, MAFF, MALAT1, MAP2K3, MKNK2, MT-ND2, MT-RNR1, MT-RNR2, MT-TM, MTRNR2L1, NAMPT, NFKBIZ, NR4A2, NR4A3, PNRC1, PPP1R15A, PSAT1, RCAN1, RGS3, RHOB, RN7SK, RND3, SLC38A2, SQSTM1, TM4SF1, UBC, WARS, XBP1, ZFAND2A, ZFAS1.
Input evidence
Reference evidence
S05 · Sec7 first paragraph; Fig5 b-d legend; reference/paper_result.txt; private answer_key.json field shared_genes
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
S06 · Fig.5b rows listed in reference/core_human_evidence.tsv; private answer_key.json and reference/baseline_audit.json
All 19 core genes have positive changes and adjusted P<0.05 in the authors' full gene table. The robust subset and VSD-shift means are recomputed from the packet; the paper does not report these new summary numbers.
Recorded result / GPT 6 astra
The scorer rejects the six-gene set. It contains six of the 19 core genes, with no extras, giving a set overlap of 0.3158 against the required 0.8. Missing genes are BTG1, DNAJB9, FOSB, GEM, HERPUD1, HSPA1A, HSPA5, IER3, IL11, INHBA, PTGS2, SAT1, and TGIF1.
Measure the mean expression shift for neuroactive drugs
Earns credit
Report the mean across selected genes of neuroactive-drug expression minus negative-control expression, in variance-stabilized units. In the submitted JSON, nad_mean_shift must be approximately 1.24028, with an absolute tolerance of ±0.2553 (inclusive). The displayed reference is rounded for readability; the original contract retains the exact value.
Does not earn credit
No credit for a missing field, a value that is not a finite JSON number, or a result outside the allowed tolerance.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph; Fig5 b-d legend; reference/paper_result.txt; private answer_key.json field nad_mean_shift
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
S06 · Fig.5b rows listed in reference/core_human_evidence.tsv; private answer_key.json and reference/baseline_audit.json
All 19 core genes have positive changes and adjusted P<0.05 in the authors' full gene table. The robust subset and VSD-shift means are recomputed from the packet; the paper does not report these new summary numbers.
Recorded result / GPT 6 astra
The scorer rejects NAD 1.5232: its difference from 1.2402766391 is 0.2829233609, exceeding the allowed 0.2553. The submitted value is correctly calculated for the submitted six genes, but does not pass this criterion's numeric rule.
Measure the mean expression shift for oncology drugs
Earns credit
Report the mean across selected genes of oncology-drug expression minus negative-control expression, in variance-stabilized units. In the submitted JSON, oncd_mean_shift must be approximately 0.211442, with an absolute tolerance of ±0.1261 (inclusive). The displayed reference is rounded for readability; the original contract retains the exact value.
Does not earn credit
No credit for a missing field, a value that is not a finite JSON number, or a result outside the allowed tolerance.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph; Fig5 b-d legend; reference/paper_result.txt; private answer_key.json field oncd_mean_shift
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
S06 · Fig.5b rows listed in reference/core_human_evidence.tsv; private answer_key.json and reference/baseline_audit.json
All 19 core genes have positive changes and adjusted P<0.05 in the authors' full gene table. The robust subset and VSD-shift means are recomputed from the packet; the paper does not report these new summary numbers.
Recorded result / GPT 6 astra
The scorer rejects ONCD 0.3913: its difference from 0.2114417685 is 0.1798582315, exceeding the allowed 0.1261. The arithmetic is correct for the submitted six genes, but the value fails the authored numeric rule.
Measure the mean expression shift for temozolomide
Earns credit
Report the mean across selected genes of temozolomide expression minus negative-control expression, in variance-stabilized units. In the submitted JSON, temozolomide_mean_shift must be approximately -0.00609948, with an absolute tolerance of ±0.0282 (inclusive). The displayed reference is rounded for readability; the original contract retains the exact value.
Does not earn credit
No credit for a missing field, a value that is not a finite JSON number, or a result outside the allowed tolerance.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph; Fig5 b-d legend; reference/paper_result.txt; private answer_key.json field temozolomide_mean_shift
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
S06 · Fig.5b rows listed in reference/core_human_evidence.tsv; private answer_key.json and reference/baseline_audit.json
All 19 core genes have positive changes and adjusted P<0.05 in the authors' full gene table. The robust subset and VSD-shift means are recomputed from the packet; the paper does not report these new summary numbers.
Recorded result / GPT 6 astra
The scorer rejects temozolomide −0.0369: its difference from −0.0060994814 is 0.0308005186, exceeding the allowed 0.0282. The arithmetic is correct for the submitted six genes, but the value fails the authored numeric rule.
Connect the selected genes to a response shared across active drugs
Earns credit
C-001 passes, and the response is justified using real positive shifts in the selected genes across multiple NAD drugs, with concrete examples from at least two drugs and at least two selected genes. Gives a correct description of the shared early program that those selected genes support, connecting the observed changes to a response shared across drugs rather than an effect of only one drug.
Does not earn credit
C-001 fails or is missing; or the submission only names a program, provides examples from fewer than two drugs or two selected genes, uses the wrong time or mismatched reference groups, gives incorrect changes, or does not connect the observations to a supported shared early program. An unresolved statement contradicting the required inference also fails this point.
Accepted alternatives
An AP-1/immediate-early response with BTG1 is one accepted description. A combined immediate-early and stress-response description is equally acceptable when supported by the selected genes, including unfolded-protein or integrated-stress genes where appropriate. No particular gene or program label is required; the description must fit the genes and observed changes. Equivalent numeric contrasts or correct clear direction-of-change comparisons using named drugs and selected genes. Clearly labeled discovery comparisons to either pooled NEG or DMSO are valid; numeric answer fields still require pooled NEG.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph, sentences on the common AP-1/BTG signature and eight AP-1 TFs; Fig5d
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
Recorded result / GPT 6 astra
This criterion explicitly requires C-001 to pass. That prerequisite fails because the submitted set omits most core genes. Bullets 1 and 3 describe positive responses across NAD drugs, but those observations about the incomplete set cannot satisfy the required supported-set conclusion.
Use the inactive-drug controls to bound the drug-class claim
Earns credit
C-001 passes, and the answer compares the selected response under active NADs with the small responses under quetiapine and paliperidone, relative to the same six-hour NEG reference or to DMSO. Uses this observation to conclude that the response is not common to every neuroactive drug merely because of its clinical category.
Does not earn credit
C-001 fails or is missing; or it treats inactive neuroactive drugs as oncology drugs, claims every neuroactive drug induces the program, or mentions controls without their observed contrast. An unresolved statement contradicting the required inference also fails this point.
Accepted alternatives
Correct named-drug contrasts using the submitted supported set; DMSO-only comparisons are acceptable for this reasoning point when clearly labeled. The numeric answer fields still require pooled NEG.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph; Fig5b,d and Extended Data Fig6 title and legend
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
Recorded result / GPT 6 astra
C-001 passing is an explicit prerequisite, and it fails. Bullet 4 reports small paliperidone and quetiapine responses, but the comparison uses the incomplete six-gene set. It therefore cannot earn the required conclusion about controls for an accepted shared response.
Distinguish the oncology group pattern from individual exceptions
Earns credit
C-001 passes, and the answer uses the selected-set shifts to explain that the ONCD group has a weaker average response than NAD and temozolomide is near zero, while noting an individual oncology exception such as osimertinib with strong FOS/JUNB induction or a selected-set mean near 0.7581 for the core set (or its correctly computed value for another accepted set). It does not claim that all oncology drugs have zero response.
Does not earn credit
C-001 fails or is missing; or the answer uses another gene set or mismatched controls for the oncology comparison, ignores the individual exception, or claims no oncology drug induces any of these genes. An unresolved statement contradicting the required inference also fails this point.
Accepted alternatives
An equally specific contrast showing osimertinib FOS and JUNB induction and weaker average ONCD response. A consistently applied, clearly labeled DMSO reference can support this reasoning comparison; numeric answer fields still require pooled NEG.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph, oncology contrast; Fig5d OSIMERTINIB and TEMOZOLOMIDE rows
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
Recorded result / GPT 6 astra
The explicit C-001 prerequisite fails. Bullet 4 correctly recognizes a weaker ONCD average, little temozolomide response, and an osimertinib exception for its six genes. However, the criterion requires those comparisons for an accepted selected set, which was not supplied.
Explain why the response is selective within related gene families
Earns credit
C-001 passes, and the answer gives at least one correct within-family contrast under six-hour NAD versus NEG: an induced gene is compared with one or more related family members that are flat or decreased. Connects these observed changes to the inference that the response picks particular family members rather than raising the whole family.
Does not earn credit
C-001 fails or is missing; or the answer lacks an observed within-family contrast, compares unrelated genes as family members, gives incorrect changes, uses the wrong groups, or does not infer selectivity among family members. A claim that all related genes rise, or another unresolved statement contradicting the required inference, also fails this point.
Accepted alternatives
For example, induced ATF3 (about 1.9819) or ATF4 (1.0856) can be contrasted with decreased ATF5 (-0.2558) or ATF6B (-0.1760); induced FOS (1.1321) can be contrasted with nearly unchanged FOSL1 (-0.0399). These NAD-minus-NEG values are examples, not required genes or a required list of contrasts. Other correct within-family contrasts from the packet are accepted. Approximate values or correct clear directional comparisons are sufficient; a small change need not be called statistically significant.
Prerequisites
C-001
Input evidence
Reference evidence
S05 · Sec7 first paragraph, sentences on ATF5, ATF6B, and FOSL1; Extended Data Fig6e legend
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
Recorded result / GPT 6 astra
The explicit C-001 prerequisite fails. The submission does provide a valid family comparison: induced FOS versus nearly unchanged FOSL1, followed by an explanation that whole families do not rise uniformly. This otherwise useful reasoning cannot overcome the criterion's required accepted gene set.
Separate pooled group shifts from drug-specific responses
Earns credit
Shows, with correct six-hour values from the packet, at least one gene whose pooled NAD-minus-NEG shift is weak but whose response to a specific drug versus NEG is strong. Compares the pooled and drug-specific values for that same gene and infers that a weak pooled response does not mean the gene never responds.
Does not earn credit
The answer only names genes or drugs, lacks correct pooled and drug-specific values for the same gene, uses mismatched times or reference groups, does not show a strong drug-specific response despite a weak pooled shift, or omits the inference that pooling can hide a response to a specific drug. Incorrect claims about which drug is strongest or an unresolved statement contradicting the required inference also fail this point.
Accepted alternatives
For example, BTG1 has a larger pooled shift than BTG2 (about 1.0321 versus 0.1418), while BTG2 rises by about 0.7686 under vortioxetine versus NEG. This supports the inference that the weak pooled BTG2 response does not mean it never responds. Vortioxetine is not the strongest BTG2 inducer across all drugs. Neither these genes nor this drug is required. Any other gene and drug showing the same supported pooled-versus-drug-specific contrast are accepted. Approximate correctly computed values are sufficient. Drug-specific comparisons in log2 ratios of mean VSD values, if clearly labeled separately from the required shift statistic, are also accepted when the pooled and drug-specific observations support the same inference.
Input evidence
Reference evidence
S05 · Sec7 first paragraph, BTG1 among top 20 genes and BTG2 particularly induced by vortioxetine; Fig5d
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
Recorded result / GPT 6 astra
The required comparison is missing from reasoning.md. No gene is shown with both a weak pooled NAD shift and a strong named-drug shift, followed by the inference that pooling can hide that response. Bullet 5 instead describes strong pooled responses with negative drug exceptions. This fails independently of C-001.
Keep the expression inference separate from proof of growth inhibition
Earns credit
Connects the observed six-hour response to induction of an early gene program, but states that these expression measurements alone do not show that the induced genes cause growth inhibition. Explains that the supplied packet measures RNA responses and contains no manipulation of these genes paired with a growth readout.
Does not earn credit
It treats expression changes as proof of causal growth suppression, dismisses the measured induction because causality is untested, or gives only a generic caution without linking it to what was measured. An unresolved statement contradicting the required inference also fails this point.
Accepted alternatives
Equivalent wording distinguishing a drug-associated expression response from demonstrated necessity or sufficiency.
Input evidence
Reference evidence
S05 · Sec7 final paragraph: circumstantial evidence followed by BTG1/2 and JUN loss-of-function experiments; Fig5j,k legend
The authors report selective AP-1/BTG induction and compare drug responses. Verbatim evidence is in reference/paper_result.txt.
Recorded result / GPT 6 astra
Bullet 8 connects the observed six-hour increases to an early transcriptional response while stating that they do not establish causal growth suppression. It explains that the packet contains expression measurements and sample annotations, with no gene perturbations or growth measurements. This meets the criterion independently of the gene-set error.