|
|
|
|
Unbound parameters and dose to achieve minimum inhibitory unbound concentration at required time point
for a median 55 kg individual
Note: The background colour of input fields in the table indicates the source of the data:
In the absence of Science Cloud data, blanks or default values were used as follows:
You can specify new values ovewriting the Science Cloud data. Enter a value in the header input field of a column to set this parameter value for all compounds at once.
MMVFree is an R Shiny web application for multi-compound dose prediction of malaria drug candidates, formulating dose criteria in terms of the free (unbound) drug concentration. It predicts the dose required for the unbound plasma concentration to stay above the unbound minimum inhibitory concentration (MIC,u) for a specified duration, for both oral and long-acting injectable (LAI) routes of administration, using in vitro clearance/potency data and a volume of distribution (Vss) estimate.
The app is developed by Medicines for Malaria Venture (MMV) to support early-stage decision-making for antimalarial drug candidates.
GetType()
) was re-derived
from scratch dozens of times per parameter instead of being computed once and reused,
and a related scan instead of an indexed lookup for looking up a parameter's display
label during Science Cloud upload processing. Measured on a real 4-compound Science
Cloud upload: parameter-lookup call count dropped from 1,486 to 1,038 calls, cutting
the time spent in upload processing itself by about a third. Most of this app's
upload-processing time turned out to be spent downstream, in plot/report rendering
rather than in parameter lookups - end-to-end upload time improved more modestly
(about 10%) as a result; the remaining cost is tracked separately in #69.
xlsxjars
upgrade - worked around the same way MMVSola does, via a custom
addPictureExact()
).
formula_doseMIC()
's
practical 1e9mg search cap - but a finite (if physically absurd) dose does
mathematically exist, since coverage duration keeps growing, just very slowly, as
dose increases without bound.
formula_doseMIC_single()
now returns
Inf
for this "exceeds the cap" case, distinctly from
NA_real_
for genuinely missing/invalid PK inputs - the Oral/LAI dose
tables and dose-landscape plots now show the top dose category ("≥ 500mg"/
"≥ 1000mg", red) for the former, while a compound with real missing data (e.g. no
Vss,u measurement) keeps its existing v1.6.7 behavior: excluded from the plot and
listed in its footnote instead.
buildDoseLandscape()
) left a
grid cell uncolored whenever
formula_doseMIC()
returned
NA
for that (Vss,u, CL,u) combination - genuinely unreachable at any
dose, not a computation error (see the v1.6.6 note above). Such cells now render in
the top (red) dose category instead, matching their true meaning: "would need an
effectively infinite dose". Real compound points are unaffected - a compound with
missing or out-of-range input data is still excluded from the plot and explained in
the existing footnote, not recolored.
formula_doseKineticsShape()
(
Resources/STATE_Objects.R
) produced
NaN
whenever absorption and elimination rates coincided
(the classic one-compartment "flip-flop" case) or either rate was exactly 0 (a
zero-clearance or zero-absorption compound) - both are removable 0/0 singularities
in the closed-form expressions, now handled via their analytic limits (or, for the
flip-flop case, a tiny, physically-negligible nudge away from exact equality).
findWindow()
separately assumed concentration always eventually
decays; for a genuinely zero-elimination compound it now correctly reports
coverage as indefinite instead of failing to find a (nonexistent) offset time. Most
blank cells on the Oral/LAI dose tables remain blank for a different, expected
reason - the target coverage duration is genuinely unreachable at any realistic
dose for a very short elimination half-life, not a numerical bug - but the crashes
from these specific degenerate inputs are fixed.
\(k_e = \frac{CL_{int,u}}{Vss_u}\)
) instead
of rendered equations - but not if some other tab was clicked first. The tab-shown
handler that triggers the page's one and only MathJax typeset (see the v1.6.3 fix,
above) assumed two things were already true the moment the tab was shown: that Shiny
had finished server-rendering the User Guide iframe into the page (it hasn't yet on a
fast click - even pre-rendered, this still needs a websocket round-trip after the page
loads), and that MathJax had finished loading and initializing inside it (
MathJax.typesetPromise
exists as a stub almost as soon as MathJax starts loading, well before it's actually
ready - calling it that early silently did nothing). Visiting another tab first happened
to buy enough time for both to become true before the User Guide tab was ever shown,
masking the race. The handler now polls for both instead of assuming either, so it
works regardless of how quickly the tab is clicked.
cdn.jsdelivr.net
- the only remaining externally-hosted asset in
the guide (every other library - Bootstrap, jQuery, highlight.js, tocify - was already vendored
locally). MathJax is now vendored locally alongside the other User Guide assets, so the guide
needs no external network requests at all - one less thing that can go wrong on a deployment
with restricted outbound access, on top of the race-condition fix above.
www/Generated/
- entirely git-ignored - so a fresh deployment had no rendered copy until someone
manually re-ran
build_docs.R
(which itself requires the
bookdown
R package) on that machine. The rendered output now lives under the always-committed
www/docs/
instead, matching the sibling MMVSimLAI app, so the User Guide is
available on any deployment without requiring
bookdown
there.
For questions or feedback, please contact:
Medicines for Malaria Venture (MMV)
Route de Pré-Bois 20, 1215 Geneva 15, Switzerland
www.mmv.org
MMVFree 1.12.6 |
|
Powered by: |
|