패치 노트

Fixed

• A VHH-Fc (heavy-chain-only) construct's VH is now recognised even when its J-segment carries an engineered V-P in place of the germline W-G — the structure map no longer leaves it Unknown.
• The lone VH of a VHH construct is now drawn tilted at the same arm angle as a normal Fab's domains, standing in the Fab's own CH1 slot instead of on a vertical stub, so the hinge line matches every other arm.

What's new, on first launch after an update

• The app now pops up what changed since you last ran it, read straight from this changelog — no separate release-notes list to keep in sync. Silent on a fresh install, and silent again if nothing changed since your last launch.

Save maps: a TXT format for a copy-pasteable sequence dump

• A new TXT checkbox next to PNG/SVG saves each chosen antibody as a plain-text file — the antibody name, then each chain's code and its sequence, nothing else, ready to paste elsewhere. Checking it switches Include to "Sequences only" and clears every detail toggle, since none of them apply to a text dump.

Sequences panel: Shift range-selects domains

• Shift-clicking (or Shift-double-clicking) a domain now selects every domain between it and the last plain pick, not just the one you clicked — the usual Shift behaviour for a range. Works from either the colour bar or the residue text, so a click that lands a little off the (thin) bar still ranges correctly instead of dropping the selection.

Fixed

• A VL region ending on a J-segment with the conservative Kabat-106 V→I substitution (e.g. the JK3-derived ...FGPGTKIDIK) is now recognised, instead of the whole chain falling through as Unknown.
• Save layout now exports the legend at the position you dragged it to, instead of always writing it back at the default spot.

ADC: a light-chain/Fc-fusion Fab arm is now drawn and conjugated correctly

• A Fab arm whose CH2-bearing chain carries CL instead of CH1 (e.g. a B-body-style construct, where one arm's "light chain" is itself fused to an Fc) now gets its L-H disulfide drawn on CL's own terminal Cys, not drifting onto that chain's own unrelated hinge. A hinge-less Fd's bare EPKSC tail (its LC-pairing Cys, with no core hinge following it) is no longer silently dropped from domain segmentation, so its Hinge 1 ADC site is now offered, placeable, and styled -- direction and linker length -- exactly like a plain light chain's CL site. CL itself, when it sits on such a heavy-role chain, gets its own bead direction rotated 45 deg clockwise instead of pointing back into that chain's own hinge/Fc.

ADC dialog: Domain autocomplete, and no two sites sharing one cysteine

• A site's Domain field now autocompletes as you type, completing to the topmost remaining match with the rest pre-selected so the next keystroke overwrites it.
• A chain's cysteine can no longer be claimed by two different sites at once: a domain already picked elsewhere on the same chain (any site, any payload group) drops out of every other site's Domain dropdown for it, and freeing one (removing that site, or changing its chain/domain) makes it available again everywhere.

Sign-in required

• The app now asks for a company e-mail before it can be used. Enter it, get a 6-digit code by e-mail, type it in — no password to set or remember. A successful sign-in is remembered for 30 days, so the app keeps working fully offline within that window; past it, sign in again the same way. Existing history (Projects/Antibodies) is untouched either way — sign-in status lives separately from your saved data.
• An e-mail that isn't recognised isn't rejected outright: it files a request and the admin gets notified, so a new teammate can ask to be let in from inside the app instead of being added ahead of time.

Intro, how-to, patch notes and feedback moved to the web

• The ❓ How to use, 💬 Feedback and 🆕 What's new toolbar buttons are gone, replaced by a single 🌐 Website button. The program intro, a fuller (and now current) usage guide, this changelog, and a public question/feedback board with answers all live on the hub site now, organised by topic instead of one long page each.

Fixed

• An additional engineered mutation site is now recognised.

Antibody-drug conjugates (ADC)

• A new ADC payloads dialog (button beside Target colours) lets you build one or more payload/linker groups and pick their conjugation sites by chain and domain — CL, Hinge 1/2/3, or a detected engineered cysteine (e.g. V205C). DAR is computed from the sites you pick, never typed by hand: CL and each hinge cysteine carry one thiol apiece, an engineered site carries one, all multiplied by however many arms that chain is drawn as.
• Picking a native site (CL or Hinge 1/2/3) means that interchain disulfide was reduced to make the free thiol the payload sits on — its line is erased from the map once the beads are placed. An engineered THIOmab site never touches a native disulfide, so nothing is erased for those. Picking Hinge 1 for a heavy chain auto-adds its paired light chain's CL as a site too, since they're two ends of the same bond — otherwise the DAR would be half-counted until both were set by hand.
• Hinge 1/2/3 and CL now reflect hIgG1's actual chemistry: the light chain's CL cysteine bonds the hinge's own N-terminal cysteine (not one on CH1), and the two hinge-hinge disulfides sit in the hinge itself, above CH2 — drawn accordingly, with the hinge redrawn as a longer vertical run to carry them correctly.
• A payload sits as a small two-tone bead: a wider outer ring and a narrower inner fill, independently coloured in the dialog. The connector line filling a reduced native bond's gap takes the outer colour. Leave either colour on Auto and it's picked for you; two antibodies in the same project using the identical payload+linker combo always end up the same colour, even left on Auto, so an ADC and its naked parent read consistently across a comparison.
• A Layout tile's caption gains one line per payload group (e.g. "MMAE / mc-vc-PAB · DAR 4"), and a Layout's merged legend collapses matching payload+linker entries from different antibodies into one, ignoring each one's own DAR.
• Re-opening the dialog on an antibody that already has ADC groups offers Save (writes everything — payload, linker, sites, colours, group count — onto that antibody in place) alongside Create New ADC (forks a separately-named antibody, leaving the one you had open untouched).

Layouts — display, style and navigation

• Two independent toggles, saved with the sheet: Antibody info (each tile's chain-code and ADC caption lines) and Legend (one merged key for the whole sheet, rather than repeating it per tile).
• Label font split into three roles — Name (the tile heading), Target (a tile's antigen/Fc/scFv labels) and Legend — each its own font and size (Sans-serif/Serif/Monospace, 6–40pt). Click the Name/Target/Legend label next to the font controls to choose which one you're editing.
• Mouse-wheel zoom, keeping the point under the cursor fixed on screen; double-click resets it. Panning is now real scrollbars, since left-drag already places antibodies here.
• The merged legend can be dragged anywhere on the sheet — a rough outline of its own box follows the cursor while you drag — and Reset legend position puts it back below the grid.
• Saving a sheet as PNG/SVG no longer draws a border around an empty cell (still shown on screen, for drag-and-drop clarity), and a transparent-background PNG no longer loses the legend's own colour swatches.

Fixed

• + Antibody and a project's right-click Add antibody now prompt for a name, like New/Rename project already do, instead of always falling back to an auto-numbered one.
• A hinge-inserted (or N-terminal-only) arm scFv's antigen label now lands close to its ovals instead of reaching out at a tandem-Fab-tuned distance meant for a different layout.

Notes

• New HistoryStore.set_drug_color/get_drug_colors + auto_drug_colors (assemble.py) give ADC payload/linker combos the same project-wide colour + remembered-auto-hue system target_colors/auto_target_colors already gives antigens; a group's own stored colour still wins over a merely-remembered auto hue, only a real explicit project-wide pick overrides it.
• layoutsheet.layout() now returns a 3-tuple (adds the grid's own size alongside the sheet size) and gained transform_ops(), combine_legend(), _legend_key(); the shared transform_ops scales every field a coordinate or length lives in (including a shape's absolute points), used by both the shrink-to-fit path and the new interactive zoom.
• seqpanel.legend_layout gained an 'adc' shape (two stacked circles) alongside the pre-existing 'dot' (KiH knob) and line-sample shapes.
• assemble.build_scene gained label_style (a Layout's own font/size, independent of the antibody map's own look) and drug_colors/adc_groups params.
• Test suite grows across test_adc.py (61 checks) and the test_layout_* suites (sheet geometry, export, editor, store, tree).

A disulfide-stabilised knob-into-hole is told apart from a plain one

• S354C and Y349C now draw the inter-chain disulfide they form. Until now the pair only tinted the CH3s knob and hole, exactly as T366W and Y407V do on their own, so a KiH carrying the extra bond and one without it came out as the same picture. The bond crosses the channel between the two CH3s, high in the domains where EU 349/354 sit and clear of the knob below.
• It takes both halves, one on each chain: a lone cysteine has no partner to bond to, so the map stays a plain KiH.
• The same bond is drawn on a CH3-swapped Fab arm — a Fab whose CH1 and CL slots hold a knob/hole CH3 pair (Additional structure 4). There it sits on the side away from the hinge, where a CH3's N-terminus points once the domain is in a Fab slot, with the knob between it and the hinge end. It stops short of the variable domains, so it cannot be read as a V-to-constant bond.

Fixed

• A CH3-swapped Fab arm no longer draws a CH1-CL disulfide. Two CH3s hold together non-covalently, so nothing bonds at that position — the arm was borrowing the bond a real CH1-CL pair has. The only disulfide there now is the engineered S354C/Y349C one, when the design carries it.

Only knob-into-hole maps change: those with the two cysteines gain the bond, and a CH3-swapped arm loses one it should never have had. Detection, sequences and saved projects are untouched — this is how the map is drawn, not what it is drawn from. tests/test_kih_layout.py grows to 25 checks, covering both placements and the arm that must stay bond-free.

Every scFv is drawn where the sequence puts it

A heavy chain's arm has one slot for the module that runs into the hinge, and every chain now uses it the same way. A plain Fab puts its CH1 and CL there. A chain whose scFv is spliced between CH1 and the hinge puts that scFv there and slides its Fab one column further out. A chain that opens with an scFv instead of a Fab puts that scFv there too. Same module, same place — so the two arms of a bispecific line up instead of drifting apart, and a map can be read straight down the sequence.

• A hinge-inserted scFv (VH-CH1-scFv-Hinge-CH2-CH3) used to hang below its arm. Making room meant lifting that arm well clear of the other one, so an antibody with the insert on only one arm came out visibly lopsided; the scFv was also tilted the mirror of the Fab, so the arm kinked halfway down. The arm is now one straight limb, both arms sit level, and it reads in order: Fv, constants, inserted Fv, hinge, Fc.
• A C-terminal scFv is drawn as the Fab's constants reflected across the Fc — a quarter turn from where it sat, clockwise on the knob arm and anticlockwise on the hole arm. Its two domains now lie along the arm the way CH1 and CL do, and land on the mirror of their slots, so the four arms read as one symmetric X. Its linker from CH3 meets the domain on the side rather than end-on, where it would fork against the scFv's own linker.
• An N-terminal scFv (scFv-Hinge-CH2-CH3) now reaches CH2 through a linker and then the hinge, split by residue count like every other arm. Before it was one green linker end to end, which hid the hinge those chains actually have.
• A C-terminal scFab sits on the footprint of the Fab reflected across the Fc, on the same four slots, so the two ends of the molecule are the same size and reach. Inside it the chain sets the order: variable pair inboard, constants outboard, the VL-CL run on the lower row and VH-CH1 on the upper one. That keeps its two green runs apart — the CH3 linker lands on VL's inner side, low and near the Fc, while the long internal linker leaves CL's outer end, far out on the other side of the cluster.

The linker inside an scFv says which domain comes first

• It is drawn as an S: out of the first domain's C-terminal end, round into the channel between the pair, along that gap parallel to both, then out to the second domain's N-terminal end. The old drawing was a symmetric loop between the two ovals, which said nothing about the order the chain runs in — that was left to the (VL-VH) note under the antigen label, which is still there.
• The two V domains now sit as close as a Fab's CH1 and CL, so an scFv reads as the Fv it is rather than two domains that happen to be near each other.
• The same S carries a scFab's long internal linker, the one joining its VL-CL run to its VH-CH1 run: out of CL's C-terminal end, down the channel between the two runs, out to VH's N-terminal end.
• An engineered ds-scFv disulfide is drawn thicker than the map's other bonds, and crosses the linker at the middle of the pair. It passes behind the domains like every disulfide does, so only the few units inside the gap show — at the standard width it was easy to miss, and where it sat before it covered the end of the linker and made it look cut short.

Every map of an antibody carrying an scFv is redrawn by this release; twelve of the fourteen reference constructs change. Sequences, detection and saved projects are untouched — this is how the map is drawn, not what it is drawn from.

Layouts — several antibodies compared on one sheet

• A layout puts several of a project's antibodies side by side on a grid, so a set of constructs can be taken in at a glance instead of clicked through one at a time. Right-click a project → Add layout; it appears under a 📐 Layouts folder and opens like an antibody.
• Opening one gives it the whole workspace — the map, chains and sequences areas become the sheet, because a comparison is unreadable in a third of the window. Clicking an antibody puts everything back, split where you left it.
• Set Rows and Columns, then drag an antibody from the list onto a cell. Drag a cell onto another to move it, onto a filled one to swap them, or off the grid to take it off the sheet. The list marks what is already placed.
• Each tile shows the antibody's name centred above its map, and below it one line per chain: the Benchling code and a short tag saying what that chain is — HC, LC, or HC · knob / HC · hole for the two halves of a knob-into-hole pair. Every map is drawn at the same scale, so a bigger construct really does look bigger — fitting each tile to its own box would have hidden exactly the difference the sheet exists to show.
• While you drag, the pointer carries the antibody's name on a badge and the cursor changes, so it is clear something is being moved and what it is; the cell it would land in is outlined.
• A cell holds the antibody by reference: edit its sequences and every layout showing it follows. Target colours carry over too — a sheet is drawn with the project's antigen palette, the same one its maps use, and recolouring an antigen shows on the sheet the next time you open it. Deleting an antibody, or moving it to another project, just empties the cells that named it.
• With a layout open the History button becomes [💾 Save layout…] and writes the sheet as one PNG or SVG — format, white or transparent background, and the sheet width. It is a single artifact, so it asks for one file rather than the batch dialog's tick-lists. An empty sheet says so instead of writing a blank image, and a cell whose antibody cannot be drawn is called out first.

Select residues in the Sequences panel and copy them

• Drag across the residues in the Sequences panel to select them, then press Ctrl+C. The selection follows the line wrap, and what lands on the clipboard is the residues alone — one line, no FASTA header, no line breaks — so it pastes straight into a sequence box, BLAST, or a spreadsheet cell.
• A whole domain in one click: double-click anywhere inside a domain, or click its coloured bar, and that domain is selected end to end — across line wraps. Clicking any part of a bar takes the whole domain, not just the run on that line, so a VH or a CH3 is one click and a Ctrl+C away.
• A selection stays inside the chain it started in. Dragging on past the end clamps there rather than running into the next chain: two chains are two molecules, and joining them would produce a sequence that does not exist.
• The range is held as residue positions, not pixels, so resizing the window — which rewraps the panel — keeps the same residues selected.

A chain can have a readable name, and the Sequences panel says more

• Each chain card gains a Name box beside Code — a readable name such as Heavy chain or anti-CD3 scFv arm. Optional, stored with the antibody, and shown in the Sequences panel so a map read back weeks later says what each chain was meant to be.
• The Sequences panel now heads each chain with two lines: the code, the name and the antigens, then the chain's domain architecture (VH-CH1-Hinge-CH2-CH3). The architecture gets its own line because an scFv or scFab arm makes it long enough to push the antigens off the edge.
• The card's head is now two rows — Code and Name on the first, Plasmid and Targets on the second — so none of the four boxes is squeezed.

Fixed

• The Save maps list only scrolled from its scrollbar. The wheel did nothing over the list itself — Tk sends a wheel event to the widget under the pointer and never up to the canvas behind it, so every entry was a dead spot. The same thing the chain cards had fixed in 1.08, now applied to that list.

The History right-click menu names its shortcuts

• Right-clicking a project or an antibody now shows the key beside each entry — Rename F2, Duplicate Ctrl+D, Delete Del — so the shortcuts are discoverable from the menu instead of only from How to use. The keys themselves are unchanged.

Notes

• A layout is a layouts list on a project, additive like the rest, so SCHEMA_VERSION stays at 2 and a file written before this reads back with none. duplicate_project remaps a copied layout onto the copied antibodies, which get fresh ids and would otherwise leave the copy's sheet pointing into the original. New layoutsheet.py holds the grid geometry as renderer-agnostic ops, the same shape seqpanel uses, and draw_scene gained a box parameter that draws a scene as one tile — with box=None every path is what it was. New tests/test_layout_store.py (27), test_layout_tree.py (23), test_layout_sheet.py (37), test_layout_editor.py (27) and test_layout_export.py (19).
• Exporting a sheet reuses the per-scene renderers rather than re-deriving the drawing: PNG renders each tile through the same render a single map goes through and composites it, SVG nests each finished map in a <g transform>. Rendering an antibody a second way for the sheet is what would let the two drift apart. scene_span moved to render_svg, beside _bbox, so all three renderers size a tile identically.
• The panel is a Canvas, so there is no native text selection to lean on. The shared layout now tags each wrapped residue line with its chain and starting residue (a residues key on that op), which lets the Tk renderer return a hit-map from pixel to residue. layout is shared with the PNG and SVG exporters; both read only the standard op keys and are unaffected — a test pins that.
• The selection band is a fixed #b3d4fc rather than a theme accent: this panel's paper is white in every palette, so one tint works throughout, while the dark palette's accent is a dark slate that would swamp the residues. The residue text is raised back over the band, which also keeps the band visible across a mutation highlight (itself a rect behind the residues).
• The band is exactly one residue cell tall, the same box a mutation highlight uses, so it stops above the domain bar and never hides that colour. The bars carry their domain's full span (a domain key on the bar op), which is what lets a click on any run of a bar select the domain whole.
• Ctrl+C is bound on the panel, not globally, so the chain cards' own copy still works. New tests/test_seq_select.py (30 checks).
• The chain name is an additive chain_name key on each stored chain, so SCHEMA_VERSION stays at 2 and an antibody saved before this reads as blank. Unlike the plasmid code it does travel into the scene, since the panel draws it. New tests/test_chain_name.py (19 checks); test_chains_wheel now walks every row of a card's head rather than assuming there is one.

Undo and redo for an action that replaced the whole input

• [Load example], opening a FASTA/JSON or a .docx project, and [Reset] each replace every chain at once. [↶ Undo] in the toolbar (or Ctrl+Z) now takes any of them back — the cards, the map, and the antibody as History holds it. [↷ Redo] (Ctrl+Y, or Ctrl+Shift+Z) puts it back again, for when you undo one step too many.
• Restoring the stored record is the point. Generate writes the input to history.json as it draws, so by the time you notice the mistake the entry on disk already holds the replacement; refilling the cards alone would leave the overwritten record there. Undo puts the chains, title and name back too, so the screen, the History panel and the file agree again.
• The two arrows sit to the right of [🆕 What's new] and grey out when there is nothing to undo or redo, so the toolbar says what is available.
• The last 10 steps are kept each way, and starting a new action drops the redo chain. Typing in a card is not undone, and neither is deleting from History — that already asks first.

Load example no longer overwrites the antibody you are viewing

• Reported 2026-08-05. With an antibody open, one misclick on [Load example] filled the cards with the Herceptin example and Generate wrote it over that antibody — in history.json as well, so the original was gone for good.
• [Load example] now adds the example as an antibody of its own under the current project and opens that, leaving the entry you were on untouched. No prompt to click through, because nothing is at risk.
• Opening a FASTA/JSON or picking a .docx project still replaces the current input — that is usually what you meant — but it now asks first when the cards have content, and is undoable either way. Declining a .docx project also puts the picker back on the project that is actually loaded.

A chain can carry its DNA plasmid code

• Each chain card has a Plasmid box between Code and Targets, for that chain's DNA plasmid code (e.g. PL0042). It is optional: the box shows a grey PL0000 hint while empty, and takes whatever you type — a code from another scheme is not rejected.
• The code is stored with the antibody in History and comes back when you reopen it, undo and redo included. The structure map and the sequence panel are unchanged, so saved images look exactly as before — this is for the record, not for the figure.

Notes

• The plasmid code is an additive plasmid key on each stored chain, so SCHEMA_VERSION stays at 2 and an antibody saved before this simply reads as blank. New tests/test_plasmid_code.py (14 checks) covers the round trip through the cards, the store and history.json on disk, the grey hint never being saved as a value, a card kept alive by a plasmid code alone, and the map and sequence panel staying clear of it.
• A restore point holds the cards, the active project/antibody and a copy of the project's whole antibody list. Snapshotting the list rather than just the open antibody is what makes undo and redo the same operation in opposite directions: it carries an antibody an action added or removed with its id intact, so redoing [Load example] reopens the very entry undo had removed. Restoring deliberately does not go through _activate_antibody: that captures the screen into the store first, which would save the replacement over the very record being restored. A theme switch clears both stacks, since their cards are about to be destroyed.
• New tests/test_undo.py (34 checks): the new-antibody path, undo of both the create and the overwrite case, redo including the id round-trip, the redo chain being dropped by a new action, the stack bound, the empty-stack no-ops, the toolbar buttons' enabled state and the confirmation's three outcomes. It drives the real key paths with event_generate, not just the handlers.
• How to use gains the Ctrl+Z shortcut under History, and its [Load example] paragraph now says the example arrives as its own antibody.

Fixed

• An appended scFv's VL went missing when its light chain came from a mouse. On a heavy chain carrying an scFv after CH3, the VL was left unlabelled — it fell through as a single long Unknown block together with the linker in front of it, and because an scFv needs two V domains joined by a linker, the whole scFv failed to group too. The map read VH-CH1-Hinge-CH2-CH3-VH, one domain short.
• The cause was the VL J-segment motif, which a variable domain is located by. It was written around the human JK germlines and was a residue-class too narrow at two positions: Kabat-107 allowed only I/V, but the murine JK5 that a chimeric/humanised light chain brings along ends ...FGAGTKLELK; and Kabat-103 allowed only K/E/Q, so the human JK5 ...FGQGTRLEIK missed as well. Both positions now take the wider class, which covers the human and murine JK sets and lambda.

Notes

• The change is one regex in domains.py. Re-running it over every sequence in db/ and reference/ (114) moved no existing match, and the constant-region references stay free of false hits, so nothing that was detected before is detected differently.
• New tests/test_vl_j_motif.py (21 checks) pins the germline JK/lambda set, the clean constant references, and the reported scFv-Fc chain annotating end to end as VH-CH1-Hinge-CH2-CH3-scFv.

Saved maps are captioned with the antibody's name

• Every image written by 💾 Save maps… now carries the antibody's name at the top, so a map dropped onto a slide says what it is. The name is the one in the History panel, read at save time — rename there and the next save follows.
• The on-screen map is unchanged; the caption is added on export only. A Sequences only export has no figure area, so it has no caption either.

An antigen on 'Auto' keeps one colour across its project

• An antigen left on Auto was coloured by its binding site's position in the hue cycle within one antibody, so the same antigen could come out green in one map and purple in the next — the two were no longer comparable side by side.
• A project now remembers the hue it first handed an antigen (auto_target_colors on the project) and reuses it for every antibody under it, saved images included. Colours assigned in 🎨 Target colours still win; if one of them claims a hue an antigen was already using, only that antigen moves to a free family. Switching an antigen back to Auto returns it to the hue it had before.
• A Fv whose colour is one of the six built-in families now keeps that family's own light (VL) shade instead of an arithmetically lightened one — so an automatic hue and the matching preset in Target colours draw identically.

Fixed

• A collapsed History folder springs back open. Collapsing one project and then using another re-expanded the first. _refresh_history rebuilds the tree from scratch and re-inserted every project open — and nearly everything rebuilds it (drawing a map, renaming, adding an antibody), so any collapsed folder was undone by the next action. _sync_selection did the same through see(), which opens a row's ancestors to bring it into view, re-expanding the project holding the active antibody.
• The open/closed state now lives in App._collapsed rather than in the tree widget, so it also survives a theme switch (which destroys and rebuilds the window) and is written to state.json on quit — the panel comes back the way you left it after a restart.

Notes

• History gains a per-project auto_target_colors map, and state.json a collapsed_projects list. Both are additive: existing files load unchanged (the keys are filled in on read) and SCHEMA_VERSION stays at 2.
• New tests/test_target_colors.py (20 checks) covers the hue bookkeeping, the colour staying put across two antibodies whose binding sites are in opposite order, and the name reaching the saved file. New tests/test_history_tree.py (19 checks) drives the folder states, the theme switch and a relaunch. out/save_full.png, out/save_map_only.png and out/save_target_colours.png are sample exports.

Chains panel scrolls from anywhere

• The mouse wheel now scrolls the chain list from any part of a card — the Code / Targets row, the labels, the entries, the ✕ button. Previously only the card's ~10px side margins answered the wheel, because Tk sends a wheel event to the widget under the pointer and never up to its parent.
• Over the sequence box the wheel still scrolls the sequence itself; once that box is already at its top or bottom, the wheel moves the card list on instead of stopping there.

Knob-into-hole always draws knob-left

• A KiH Fc is now always drawn knob on the left, hole on the right — the way KiH designs are conventionally drawn — no matter which heavy chain you entered first. Previously the arms followed the input order, so entering the hole chain first put the knob on the right.
• The whole arm moves with its chain (its Fab / scFv, antigen labels, disulfides and linkers), and light chains still pair with the heavy chain they were entered alongside. The sequence panel keeps the order you entered.
• Nothing changes for a plain IgG, for a KiH entered knob-first, or for an Fc that is not a knob/hole pair.

scFv orientation on the map

• An scFv binding site's antigen label gains a second line giving its variable-domain order in N→C direction — (VL-VH) or (VH-VL) — so a single-chain arm's orientation is readable off the map. Fab, scFab and single-domain sites are unchanged (they carry no orientation note).

Chains panel

• Each chain card shows its #N position next to the Code field, renumbered as cards are added or removed — so a chain stays identifiable while scrolling a long input list.

Colour themes

• Pick a colour theme (🎨 Theme, top-right of the toolbar): five palettes — Neutral blue (default), Warm indigo, Teal, Cool slate, and Graphite (dark). The whole app repaints at once and the choice is remembered across restarts. The structure map and sequence panel keep their own colours, so PNG/SVG exports look the same under every theme.
• Fresh flat styling throughout — the app now uses the clam ttk theme with a single accent, hairline borders, roomier padding, a proper UI font (Segoe UI / Malgun Gothic where available), and a real ✓ checkbox indicator.

Save maps: choose what to include

• The 💾 Save maps… dialog gains an Include section: a preset — Everything, Structure map only, or Sequences only — over six detail toggles (domain labels, Anti-… target labels, Fc status text, legend box, sequence colour bars, mutation highlights). Toggles not relevant to the chosen preset are disabled.
• Structure map only exports are saved on a transparent background (PNG and SVG), so the map drops straight onto a slide.
• Filenames now record the content: project_antibody_content, where content is full / map / seq plus a no-… piece for any detail switched off (e.g. …_map_no-legend). Saving again never overwrites — a name that already exists in the folder gets _2, _3, …

Fixed

• Batch export no longer silently overwrites a file saved earlier with different options; existing files in the target folder are detected and the new file is numbered instead.

Target colours

• Assign a colour per antigen (🎨 Target colours, above the map): the antigen's VH oval and its Anti-… label take the chosen shade, its paired VL a lighter shade of the same hue, and the sequence-panel domain bars follow. Pick from 12 presets or any custom colour.
• The assignment belongs to the project, so every antibody added under it is drawn with the same target colours — including batch Save maps… exports. Duplicating a project copies its colours.
• Antigens left on Auto keep the automatic per-clone hue, but skip any family already assigned to another target, so two targets never share a colour.

Tandem-Fab (BiXAB) support

• New architecture: a tandem-Fab heavy chain — two Fab modules in series on one chain before the Fc (VH-CH1-linker-VH-CH1-Hinge-CH2-CH3), e.g. the Va7.2 × CD19 BiXAB. The map draws both Fabs stacked along each arm (the N-terminal Fab at the tip, the C-terminal Fab next to the hinge), each paired with its cognate light chain matched by antigen target.
• Non-canonical kappa J now recognised: a light chain whose J region carries E/Q at the conserved Kabat-103 position (e.g. FGGGTELEIK) is detected as a VL instead of falling to unknown.
• The cysteine-free connector joining the two Fabs (a hinge-derived linker whose CPPC cysteines are mutated away) is drawn as a linker, not a disulfide hinge, and drops out of the structural architecture.

History: multi-select, shortcuts, project duplication

• Multi-select in the history tree: Ctrl/Shift-click to pick several projects and/or antibodies at once, then 💾 Save maps… (or right-click → Save selected maps…) exports them all together — a selected project contributes all of its antibodies. Building a multi-selection no longer reloads the map; single-click still opens an antibody as before.
• Keyboard shortcuts on the selected row: F2 = rename, Del = delete (with confirmation), Ctrl+D = duplicate.
• Duplicate a whole project (Ctrl+D on a project, or right-click → Duplicate): deep-copies the project and every antibody under it, with fresh IDs, placed just below the original.

Structure map

• Fc effector-status tag to the right of the Fc: wild type (grey) when the Fc carries no effector mutation, or the specific Fc-null mutations in EU order (e.g. L234A, L235A, N297A, red) — LALA, N297x, P329x, GRLR, FES, etc.

Layout & inputs

• History + Structure map together now span the same width as the Chains column (Chains takes half the window); History keeps its width and the map absorbs the change. The split holds on resize.
• Sequences panel reflows its residues-per-line to fill the panel width (no right-side whitespace). PNG/SVG exports stay at the fixed 50 per line.
• Chain sequence input boxes are shorter (2 rows) at full card width.
• Generate map ▶ moved from the toolbar into the Chains action row, equal width beside ＋ Add chain and Reset.
• Removed the redundant PNG / SVG / Fit buttons under the map — save via the history Save maps…, fit by double-clicking the map. That freed strip is now part of the status log.

History: Projects → Antibodies

• The history panel is now a two-level Project → Antibody hierarchy (replacing the flat entry + category model). Create a project (＋ Project), then add antibodies under it (＋ Antibody); a new antibody is named after its first chain code when you Generate. A blank project name auto-numbers (Project N).
• Right-click a project → Rename / Add antibody / Save maps… / Delete. Right-click an antibody → Rename / Duplicate / Move to (another project) / Save map… / Delete. Deletes ask for confirmation.
• The active project and antibody are shown bold with a tinted background; each antibody row shows its target, each project an antibody count.
• List order is stable (creation order, no bubbling); the last-open antibody is restored on launch (recorded in state.json). The window opens maximized so it fits above the taskbar.

Batch export

• Save maps… (history-panel button, or right-click a project/antibody): pick any projects and/or individual antibodies and export every map to PNG and/or SVG into a chosen folder in one go.

Fixed / added (folds in the unreleased 1.02.1)

• Update install could fail while closing the running app. The installer now force-closes the running instance in PrepareToInstall (Restart Manager's interactive close is disabled), and the app quits itself right after launching the installer, so absmap.exe is never locked during the update. The post-install launch runs as the original (non-elevated) user (runasoriginaluser).
• "What's new" popup on the first run after an update, and the 🆕 What's new toolbar button.

History panel & categorization

• Left-hand history (feedback N7): the app now keeps a persistent, named list of maps. Each entry is a document — the chains you typed plus a free-text category — created first (＋ New entry), then drawn. Click an entry to reopen its inputs and map; rename/recategorize or delete from the panel. Entries are grouped by category and survive app restarts.
• Inputs (including sequences) are stored locally only, under the per-user app-data dir (%APPDATA%\absmap\history.json on Windows), and are never sent anywhere — the hub client does not read this file.

Language

• The GUI is now fully English (toolbar, panels, the How-to-use guide, the feedback dialog and its categories, update prompts, and all messages).

Feedback & updates

• In-app feedback (💬 button): pick a category, describe the issue, and optionally attach the current map — rendered without the sequence panel, so no amino-acid sequence ever leaves the machine. Falls back to a pre-filled email if the hub is unreachable.
• Update check on startup: a non-blocking banner appears when a newer build is available; the app downloads the signed installer directly (sha256-verified) and you install it — no silent auto-update.

Prediction engine

• Domain architecture predicted from amino-acid sequence alone — no alignment tools or third-party dependencies (pure standard library).
• Variable domains (VH/VL) found via conserved J-segment motifs; constant regions (CH1/CH2/CH3/CL) located by ungapped scan against wild-type human IgG1/kappa references, which also extracts point mutations.
• Automatic detection of scFv grouping, chain pairing, hinges and linkers.
• Engineering features recognised: knob-into-hole (KiH), Fc-null variants (LALA, N297A, and related), disulfide-stabilised scFv, single-chain Fab (scFab), CrossMab/CH3-crossover, C-terminal appended domains, Fc-only (one-armed) partners, and the P-O-T (obscurin/titin) light-chain-pairing scaffold.

Structure map

• 2D cartoon in the BIOINTRON construct-diagram style; each binding site (Fv) gets its own colour family (VH dark / VL light).
• Per-Fv antigen labels (Anti-…) from user-supplied targets.
• Sequence panel under the map: per-chain sequence with coloured domain tracks and background highlights for engineering mutations.
• Legend box keying the disulfide / hinge / linker connectors.
• Export to SVG (vector) and PNG.

Desktop app (GUI)

• Chain-card input (Code · Targets N→C · Sequence) — no hand-written FASTA.
• Full-height map pane with mouse-wheel zoom (0.5×–12×) and drag-to-pan; double-click / Fit resets the view.
• Load example fills in trastuzumab (anti-HER2 IgG1), a public construct.

Packaging

• Standalone Windows executable (absmap.exe, no Python needed on the target PC) built via a local Wine cross-build.
• Inno Setup installer (absmap-setup.exe) with Start-menu / desktop shortcuts and an uninstaller.
• Application icon and internal-use license bundled.

Verification

• 42/42 reference chains match the documented architecture (python tests/verify_phase1.py).