Lodestar

1. Getting Lodestar

The web app, the iPhone, iPad and Mac app, and your account.

The web app

Lodestar runs at https://lodestar.skia.bio. Sign in with the email and password your admin gave you; use Forgot your password? on the sign-in page if you were never sent one.
The web app is where runs are started and reviewed, assets are given a call, and admins manage users. Everything the app shows is there too.

The app for iPhone, iPad and Mac

Lodestar is distributed through the App Store as an unlisted app: it opens from a link and does not appear in search. Ask your Lodestar admin for the link.
It needs iOS or iPadOS 26 or later, or macOS 26 or later.
Sign in with the same email and password as the web app. If asked for a server address, enter https://lodestar.skia.bio.
The app talks to the same server as the web app: runs, decisions and watches you make in one appear in the other.

Accounts and roles

Accounts are created by a Lodestar admin; there is no self sign-up.
Every account is a viewer, a reviewer or an admin (see Roles below). Your role is shown under your name in the web app's header and in the app's Settings.
Devices you have signed in on are listed under Profile → API tokens on the web, where a lost device can be signed out.

2. Getting started

Sign in and find your way around.

Use the same email and password as the Lodestar web app. The app keeps you signed in until you log out.
If your team runs its own server, tap Server settings… on the sign-in screen and enter its address (it must start with https://).
Turn on Use demo data to explore the app with sample runs. Any email and password will work in demo mode, and nothing is sent to a server.

The layout

On iPad and Mac the window has two columns: the runs browser on the left and the current screen on the right.
On iPhone the same areas are tabs: Dashboard, Runs and Portfolio.
Portfolio (top of the browser, or its own tab) lists every asset the funnel has concluded on, across finished runs.
Your account and settings are behind the gear at the bottom of the browser (iPad/Mac) or the top of the Runs tab (iPhone).
Focused tasks — starting a run, reviewing, settings — open in their own window on top and close with Cancel or Done.

3. Dashboard

What's running, what's finished and what needs a look.

Reading the tiles

Runs — everything in the current scope (see Mine / Others / All below).
Working — runs still in progress. They update every few seconds.
Ready — finished runs with a shortlist to review.
Need attention — runs that failed, finished with no results, or produced an incomplete shortlist.
Tap any tile to see the runs behind it; tap a run to open it.

Below the tiles

In progress shows live progress bars for running jobs.
Needs attention lists the reason for each problem run so you can decide whether to re-run.
Recent results are your latest finished shortlists; Runs by input mode shows which entry routes your team uses most (tap a bar to filter the browser).

4. Finding runs

Scope, search and filter the list of runs.

Mine / Others / All

Mine shows runs you started; Others shows your colleagues' runs; All shows everything on the server.
The choice applies to the browser and the dashboard together, and is remembered next time you open the app.

Narrowing the list

Type in the search field to match run names, ids (#33), who started them, input modes or requested indications.
Use the chips (Working, Ready, Failed, Pinned) for a quick status filter, or the filter menu for input mode.
Runs in progress are grouped at the top; pinned runs above those.
Choose Clear filters from the filter menu to reset everything.

Getting back

Use the Back and Forward arrows at the top of the right-hand column (⌘[ and ⌘] on Mac) to retrace your steps.
Pull down on any list to refresh, or use the Refresh button (⌘R on Mac). Working runs refresh themselves.

4a. Projects

A folder per counterparty or asset.

Projects (in the top bar) lists every folder with how many runs it holds, and the unfiled runs. Name a project the way the deal tracker names it: the company or the asset.
Open a project to see each asset once across all of its runs — the latest verdict, every reviewer's call, and which runs it appeared in — so an asset that has been run three times under different inputs has one history. Below that, the runs themselves.
When you start a run, pick its project on the form or type a new project name there; the next run from the same browser starts in the same folder. "New run here" on a project page does the same.
A run can be filed or moved at any time from the project picker at the top of its page, or filed from the project page's "File an existing run here". Unfiled runs behave exactly as before.
Rename or archive a project from its page. An archived project is hidden from the lists and pickers until you show archived; its runs stay put.

5. Starting a run

Feed compounds through the funnel to a ranked shortlist.

Tap + (New run) in the browser or on the dashboard. On Mac, press ⌘N.
Optionally give the run a label. You can always rename it later.
Choose how compounds enter the funnel (see below) and fill in that mode's fields.
Tap Run funnel (⌘↩). The run opens straight away and shows its progress.

Input modes

Bring your own — paste known drugs, one per line, by name, SMILES, InChIKey or ChEMBL id. Import .txt / .csv… reads a file's first column.
Known target — a proprietary compound where you supply the gene target, optionally its action, modality and compartment, other names it goes by, and indications to always evaluate. Client-supplied safety and IP data must be complete to be accepted; a patent needs its expiry or filing date.
Where the asset failed (known target) — the indication, phase, reason and the sponsor's words, plus the trial id so its posted results are read. The failed indication and its relatives are never proposed; a safety reason is a pass.
Start from a target — name a gene (LOXL2, TYK2) or an Ensembl id; Lodestar lists every drug and clinical candidate recorded against it, you untick any you do not want, and each runs as a known-target compound. Biologics are listed but not run. This is what to use when you have a target in mind rather than a molecule.
Start from an indication — name a disease; Lodestar finds compounds hitting its targets and assesses them against that disease only.
Screen failed trials — terminated and withdrawn Phase 2/3 programmes from ClinicalTrials.gov, resolved, enriched and screened before pairing. Choose phases, trials per page (25, 50 or 100) and 1–5 pages. Start small: every 100 trials is about forty small molecules.
Source from ChEMBL — sample candidates by highest development phase and how many you want.

Check before you run

When you press Run funnel, every entry is resolved first. If all of it can run, the run starts at once. If anything cannot — a gene symbol typed where a compound goes, a target that is not a gene, a name nothing matches — a check panel shows what each entry resolved to and, for the rest, why not and what to do instead. Nothing starts until you press Start run.
A gene symbol typed as a compound is named as such, with a one-click Start from target button that switches to a target-first run on it.
A target-first run always stops at the check panel, so you can choose which of the listed molecules to run.
On the API, POST /runs/preview gives the same check; POST /runs refuses a submission nothing in which can run.

Advanced

Risk posture sets how much novelty the run looks for. De-risked puts clinical precedent first: indications where programmes against the target have already reached the clinic rank up, and novelty counts for less. White space puts novel biology first: indications with human genetics or in-vivo evidence and no programme against the target rank up, and crowded pairs are discounted. Balanced, the default, is the standard weighting. The posture changes which indications are proposed as well as how they score, so a run keeps the posture it started with; to compare, start a second run. A run under a posture other than Balanced says so on its page.
Secondary targets also pairs each compound through up to three other proteins it hits at 100 nM or better. It is on by default for Bring your own and Known target; pairings reached that way are tagged.
Custom scoring weights change how the eleven dimensions combine, the same way the web wizard's weights do. Neutral is 1; 0 ignores a dimension.
Model spend is capped per run and per month. When the month is spent, starting a run is refused with the reason.

Run again

Right-click (or long-press) a Bring-your-own run and choose Run again… to start a new run pre-filled with the same compounds.

6. Following a run

Runs take a minute or several.

While a run works, its Overview shows the current step, a progress bar and, once the pace is known, a time estimate.
The browser row and dashboard tile update as it goes; you can leave and come back.
When it finishes, the Shortlist tab appears and the Review button lights up in the toolbar.
A yellow Incomplete badge means a data source was unavailable for part of the run — the shortlist is usable but may be missing pairings.
No results means none of the compounds could be paired with a disease; the Compounds tab explains which inputs couldn't be used and why.

7. Reading the shortlist

What the scores and tags mean.

Each pairing

A pairing is one compound proposed for one indication. Rank and the ring score (0–100) come from Lodestar's scoring; unranked pairings could not be scored.
Tags show its review state, whether in vivo evidence exists, and whether the compound's direction of action is aligned with the disease.
Open a pairing to see the score broken down by dimension, the rationale with its citations, safety and direction notes, and the evidence summary.
Tap the compound name to see the asset: the funnel's verdict and its reasons, what its failed trial showed, safety, target biology, IP and who holds it. Client-supplied data is marked as not independently verified.
A Secondary target tag means the pairing was reached through another protein the compound hits, not its primary target.

Working the list

Search by indication, compound or target; sort by rank, score, indication or compound.
The Pending / Accepted / Rejected chips show how much review is left.
On iPad and Mac the pairing opens beside the list so you can move through it quickly; on iPhone it opens on its own screen.

8. Two scores: Lodestar and deal

Every pairing carries both; the run decides which leads.

The Lodestar score (gold) combines the eleven dimensions. The deal score (navy) is the deal rubric: five criteria — graded evidence, Phase 3 simplicity, Phase 1 safety, IP runway and the competitive race — each 0–100 with a line explaining it.
A reviewer chooses which score a run is read by on the web. The shortlist opens in that order, its rows lead with that rank and score, and Run details says Ranked by. Sort by the other rank any time.
An unknown criterion counts as 50, so Coverage on the pairing says how much of the deal score is actually measured.
Oncology indications cannot make an asset an Acquire; Orphan eligible and the number of Phase 3 programmes racing in the indication are shown beside the rubric.
Trial counts and racing programmes come from ClinicalTrials.gov and the two European registries (CTIS and EudraCT) read together, each sponsor's programme counted once. A Phase 3 running only in Germany counts. Australia and Asia are still a gap: their registries have no machine access.

IP on the pair

Besides the compound's own patent position, each pairing that passes triage is checked for method-of-use patents: does anyone claim this compound, or its target class, in this indication? The reading sits under IP on the compound page's dimension detail, and in the deal rubric's IP line.
A third party's use patent on the pair cuts the IP score hard: the thesis would run into it, unless the holder is the seller or will license. A claim over the target class cuts less. The sponsor's own use patent protects the pair, even for a molecule with no composition-of-matter position. Nothing found means a use filing there is open, which the note says.

9. Reviewing pairings

Accept, reject or edit — decisions are shared with the whole team.

Deciding on a pairing

Open the pairing and use Accept or Reject. Your name and the time are recorded and visible to everyone.
Rejecting asks why: tick the reasons (not a real indication, mechanism points the wrong way, evidence too weak, too crowded, compound cannot reach it, not our strategy) and add a note if you like, then confirm. Each reason names a mistake the funnel can make, and the count of each is where it gets fixed.
For quick calls, swipe a row in the shortlist (iPhone/iPad) or right-click it (Mac).
Each sentence of the rationale can be accepted, rejected or reworded from its ⋯ menu. Deciding every claim also accepts the pairing.

Review session

Tap Review N in a run's toolbar to step through its pending pairings one at a time.
Use Reject, Skip and Accept at the bottom — or the keys R, → and A. ← goes back a pairing.
Finish later (Esc) keeps everything you've decided so far; the summary at the end shows what you accepted, rejected and skipped.

10. Assets: verdicts, calls, memos and watching

What the funnel concluded about a compound, and what you decide.

Reading an asset

Funnel verdict — Acquire, Consider or Pass, drawn when the run finishes, with the reasons for and against. Each reason says whether it rests on the public record, the company's dossier or the model's reading.
Not assessed — the asset was looked at and found to be nothing the funnel can judge: it did not resolve to a molecule, or nothing could be paired with it. The reason says which, and when what was typed is a gene symbol, a Start from target button offers the run that was meant. It is deliberately not a Pass: a Pass says judged and failed.
Screened out means the compound was never paired: stopped for safety, or still in active development in a failed-trials screen.
What the failed trial showed — read from the numbers the trial posted: whether the primary endpoint was missed, findings that separated anyway, and any safety signal.
Safety — Clean, Not clean, No results posted, or Approved product (on-label safety; registry counts are other programmes' trials and not held against it).
Target biology — what the target's own biology says about dosing it for years: liabilities, whether humans tolerate losing it, mouse knockouts and a concern level.
Availability — what SEC filings say: who holds the asset, whether it is seeking a partner, licensed out, returned from a former partner, acquired or discontinued, and the filings read. The status carries the date it was last known true and the filing it rests on; one more than two years old is flagged as possibly changed, and the verdict says so.

Your call

Reviewers and admins record their own Acquire / Consider / Pass beside the funnel's verdict. It is saved under your name; choose Change… to revise it. On the web, the same call can be made from the run page's verdict strip or from the compound page.
Everything beside the call is optional, and all of it teaches the model. What drove it — tick the factors that made the difference (evidence, indication fit, safety, IP, competition, commercial, stage, deal terms, strategic fit); they map onto what the funnel scores and are what the weights are set from. Your reasoning — as much as you like: what you weighed, what the funnel got right or wrong, what you would want it to know next time. A question or suggestion — for the team, or about how Lodestar read the asset; these are collected on the Feedback page and answered there, and the answer appears under your call.
The compound page lists every call on the asset with its factors, reasoning, question and answer.

Memo

Tap Memo for the decision memo: everything known about the asset on one page. Copy the markdown (⇧⌘C) or share it straight into a document; the structured data sits beneath.

Watching

Tap the eye to watch an asset. Watched assets are re-read weekly — patents, active trials, SEC filings, the race in the best indication and the verdict — and you are emailed only when something moved. Tap again to stop.

11. Portfolio

Every asset with a verdict, across finished runs.

Portfolio gathers each compound from the finished runs in the current scope (Mine / Others / All) with the funnel's verdict on it, newest run first. It reads each run once and fills in as it goes.
Tap a tile or chip to show only Acquire, Consider, Pass or assets with no verdict yet; search by asset or run name.
Tap an asset to open it; right-click (or long-press) to jump to one of the runs it appeared in.

11a. Feedback

What reviewers have told Lodestar.

Feedback (in the top bar) collects every question and suggestion left with a call, newest first, with the asset, the run, the call and the reviewer. Admins answer them there; the answer appears under the reviewer's call on the compound page. Filter Open, Answered or All.
Below the questions: how often each factor was ticked, by call, and how often each reason was given for rejecting a pairing. This is the record the model is tuned from.
The weights the calls imply. Each factor speaks to one or more of the eleven scoring dimensions; a dimension cited more often than average is weighed more than the defaults do, and the table shows each weight in force, the weight the calls imply, and the change. Once there are twenty calls with factors and three Acquires among them, an admin can adopt the proposal as the default weights for new runs (and revert to the built-in defaults at any time). Runs already scored keep the weights recorded on them.

12. Naming and pinning

Keep the list meaningful.

Rename a run from the pencil beside its title, its ⋯ menu, or by right-clicking / long-pressing it in any list. Names are saved to the server and appear for everyone; leave the name empty to clear it.
Pin runs you're working on so they sit at the top of the browser. Swipe a row or use its menu. Pins are private to this device.
Open on the web takes you to the same run, pairing or compound in the Lodestar web app.

Deleting a run (admins)

Admins can delete any finished run — ready or failed, results or not — from its ⋯ menu, by right-clicking / long-pressing it, or by swiping a row left. The app always asks first.
Deleting removes the run from the server for everyone, with its compounds, pairings, evidence, every reviewer decision on them, and any watches. It can't be undone.
A run still working can't be deleted; wait for it to finish or fail. A stuck run is marked failed within about three hours, and can then go.

13. Exporting

Take a shortlist into Excel or a report.

In a finished run, choose Export shortlist as CSV from the ⋯ menu.
The file contains every loaded pairing with both rankings (rank and score, deal rank and deal score), compound, target, any secondary target, indication, ontology id, discovery route, evidence verdict, direction, review state and which ranking the run is read by.
For one asset, the Memo's Copy markdown and Share buttons take the decision memo into a document.
Long shortlists load in pages — tap Load more at the bottom first if you want every pairing in the export.

14. Keyboard shortcuts

Mac and iPad with a keyboard.

Anywhere

⌘N — New run
⌘R — Refresh runs
⌘[ / ⌘] — Back / Forward
⌘↩ — Run funnel (in the New run window)
⇧⌘C — Copy the memo's markdown (in the Memo window)

Review session

A — Accept
R — Reject
→ — Skip ← — Previous
Esc — Finish later

15. Roles

What each account can do.

Viewers can browse runs, shortlists, pairings and compounds, and export CSVs.
Reviewers can also start runs, rename them, accept, reject or edit pairings and claims, record their call on assets, and watch them.
Admins have everything reviewers have, can delete finished runs, and manage users (invite, roles, deactivation) and the indication lookup in the web app — the API doesn't offer those.
Your role is shown under your name in the account footer and in Settings.
A sandboxed account (an App Store reviewer, a demo) works in full but sees only the runs it started itself; admins set this when inviting, or later from Users.

16. Troubleshooting

When something doesn't look right.

"Your session has ended" — your token was revoked or expired. Sign in again. Devices you've signed in on are listed under Profile → API tokens on the web, where a lost device can be revoked.
"Your role doesn't allow this" — ask an admin to change your role in the web app.
"Too many requests" — the server is rate-limiting; wait a minute and try again.
Nothing loads — check the server address in Settings. Changing the server or switching demo data on or off signs you out.
A run seems stuck — the app polls every few seconds while a run works; if the server reports a failure it appears under Need attention with the reason.
Lost your place — use Back at the top of the right-hand column, or the Dashboard entry at the top of the browser.
"This server doesn't support …" — the server is older than the app and lacks that route (memo, watch, asset decisions arrived with v3). Everything else keeps working.
"errors.budget" when starting a run — the month's model spend is used up; ask an admin.

17. Getting help

Who to ask, and what to send.

Questions, a problem with a run, or an account request: Email lodestar@skia.bio.
When reporting a problem, include the run number (#33), the compound, what you expected and what you saw, and the app or browser version. Nothing about a run is confidential to the person who started it: any reviewer can open it.
The API guide for developers is at https://lodestar.skia.bio/api-guide (signed-in users).
Lodestar's version is shown in the web app's header and in the app's Settings; this guide describes version 3.2 and later.
How Lodestar handles your information is set out in the privacy policy.