# Ganesh Chandan Kanakala — Associated Public GitHub Projects **Evidence snapshot:** 2026-08-05 **Scope:** Public repositories outside `kganeshchandan/*` with commits attributable to `kanakala.ganesh@research.iiit.ac.in`, corroborated by commit history and, where available, repository contributor files. This supplements—not replaces—the 28-repository account-owned inventory. ## Discovery and attribution method - Searched public GitHub commit attribution for the IIIT-H research email shown in repository history. - Inspected the complete Git history of every external repository found in the available result pages. - Used commit timestamps for contribution chronology. These are not claimed as GitHub repository-creation dates. - A commit count records commits matching the research email; it does not imply sole intellectual ownership of upstream methods, datasets, or papers. - Several repositories are organization copies, course adaptations, or collaborations and are labeled accordingly. ## Summary | Repository | First repository commit | Latest repository commit | Total commits | Ganesh-attributed commits | Classification | |---|---:|---:|---:|---:|---| | [`devalab/MolGPT2.0`](https://github.com/devalab/MolGPT2.0) | 2025-02-18 | 2026-07-23 | 15 | 2 | Research implementation; collaborative current repository | | [`devalab/Protein-Ligand-Dataset-Bias`](https://github.com/devalab/Protein-Ligand-Dataset-Bias) | 2022-03-24 | 2022-06-29 | 8 | 8 | Research code; entirely attributed to Ganesh in current history | | [`devalab/Spectra2Structure`](https://github.com/devalab/Spectra2Structure) | 2024-05-06 | 2024-07-06 | 2 | 2 | Publication-associated research implementation | | [`D4-course/EDDTR`](https://github.com/D4-course/EDDTR) | 2022-11-30 | 2022-12-10 | 6 | 6 | Course implementation/adaptation | | [`D4-course/Symmetry-aware-actor-critic`](https://github.com/D4-course/Symmetry-aware-actor-critic) | 2022-11-30 | 2022-12-10 | 8 | 8 | Course import/adaptation of MolGym | | [`jai-ganesh-sankar/ML4NS`](https://github.com/jai-ganesh-sankar/ML4NS) | 2022-04-13 | 2022-05-18 | 12 | 8 | Collaborative course project | | [`virtual-labs/exp-defects-in-crystals-iiith`](https://github.com/virtual-labs/exp-defects-in-crystals-iiith) | 2022-09-06 | 2026-01-28 | 17 | 4 | Virtual Labs collaboration; reciprocal-lattices experiment despite legacy repository slug | | [`virtual-labs/exp-miller-indices-iiith`](https://github.com/virtual-labs/exp-miller-indices-iiith) | 2022-09-06 | 2026-01-28 | 18 | 4 | Virtual Labs collaboration; Miller-indices experiment | | [`virtual-labs/exp-symmetries-iiith`](https://github.com/virtual-labs/exp-symmetries-iiith) | 2022-09-06 | 2026-01-28 | 32 | 16 | Virtual Labs collaboration; crystal-symmetry experiment | All timestamps in this document preserve the commit timezone recorded by Git. Dates in the table are shortened for readability. --- ## 1. devalab/MolGPT2.0 - URL: . - Current default branch: `main`. - Current HEAD observed: `1b3822357e0f203ac840e21dcc623a3dcdcebff5`. - Repository history: 15 commits from 2025-02-18 through 2026-07-23. - Ganesh-attributed history: - 2025-02-18 — `f01bd4019a8aa3ed6556824348ebb13930f36d6b`, “Initial commit,” authored as **K Ganesh chandan**. - 2025-03-04 — `69ed99df6a015d28643051f8f090d30696128f7c`, “added codebase,” authored as **Kanakala Ganesh**. - The second Ganesh commit added approximately 9,801 lines across 50 files, including: - Dockstring analysis and affinity-prediction notebooks; - preference-data creation scripts; - DPO training experiments; - docking and pose-visualization tools; - multi-property encoder–decoder implementations; - corpus/vocabulary builders; - experiment images and submission scripts. - Later contributors in 2026 added/refined datasets, MinMaxScaler/configuration handling, reusable training/generation/DPO/analysis scripts, documentation, checkpoints, and results. Therefore, describe Ganesh as the **initial repository/codebase author and research contributor**, not sole author of the current repository state. ### Repository-stated paper > “MolGPT2.0: Multi-objective molecule generation via transformer encoder-decoder model and direct preference optimization.” The repository calls this a paper but does not provide a DOI, venue, or final publication status in the README. ### Current documented workflow 1. Install dependencies, with Dockstring/Open Babel recommended through conda-forge. 2. Train encoder–decoder base models conditioned on affinity or multiple properties. 3. Sample molecules and save generation metrics. 4. Perform Direct Preference Optimization using preference data. 5. Compute generated-molecule properties and produce plots. Documented property-conditioning example: - affinity - LogP (`logps`) - synthetic accessibility (`sas`) - topological polar surface area (`tpsas`) - quantitative estimate of drug-likeness (`qeds`) ### Public data and artifacts - Dockstring training data: `data/lck_dockstring_data1.csv`. - Affinity preference data: `data/PreferenceData_affinity.pkl`. - SMILES corpus. - Base and DPO checkpoints. - Generated molecules, molecule images, KDE plots, property metrics, validity/novelty/uniqueness/diversity metrics. ### Metrics currently committed at temperature 1.0 | Checkpoint | Validity | Novelty | Uniqueness | Internal diversity | |---|---:|---:|---:|---:| | Affinity base model | 98.046875% | 94.123506% | 100% | 0.868800 | | Affinity DPO model | 96.09375% | 100% | 100% | 0.872811 | | Affinity + QED base model | 97.753906% | 97.302697% | 100% | 0.865967 | | Affinity + SAS base model | 96.972656% | 100% | 100% | 0.868420 | | Affinity + LogP base model | 97.65625% | 100% | 100% | 0.867542 | These results refine the résumé claim that validity and novelty exceed 95%: that is true for the DPO and listed multi-property checkpoints, but the committed affinity-only base checkpoint has novelty of approximately 94.12%. The portfolio should attach metrics to the applicable model/evaluation rather than make an unqualified universal claim. --- ## 2. devalab/Protein-Ligand-Dataset-Bias - URL: . - Current default branch: `master`. - Eight commits, all attributed to Ganesh's IIIT-H research email. - Contribution period: 2022-03-24 through 2022-06-29. - Closely related to `kganeshchandan/dd_code` and the ACS Omega latent-bias paper. - Contains: - DAVIS and KIBA split-generation scripts; - PDBbind folds; - DeepDTA reproduction; - modified GraphDTA scripts based on author-released code; - Pafnucy training/testing plus ligand-only and protein-only experiments; - SimCNN-DTA implementation/reproduction; - SMINA experiments; - Protein–Ligand Extended Connectivity Fingerprint experiments; - PocketMatch score calculations; - added folds for PLEC and PocketMatch. This is a stronger public organizational repository for the binding-affinity bias research than the personal `dd_code` mirror alone. --- ## 3. devalab/Spectra2Structure - URL: . - Current default branch: `master`. - Two commits, both attributed to Ganesh's IIIT-H research email. - Contributions: - 2024-05-06 — initial implementation. - 2024-07-06 — `qm9.py` update. - Associated with the Digital Discovery Spectra-to-Structure paper. - README instructs users to obtain data/splits from the paper's ESI, configure raw train/validation/test pickle paths, create the environment, run `run.py`, and log results to W&B. - A personal-namespace copy also exists at `kganeshchandan/Spectra2Structure`; treat the two URLs as copies of one research implementation, not two independent projects. --- ## 4. D4-course/EDDTR - URL: . - Six commits, all attributed to Ganesh. - Contribution period: 2022-11-30 through 2022-12-10. - Title: **Explainable Deep Drug-Target Representations for Binding Affinity Prediction**. - Explores CNN reliability and input-region explanations for binding-affinity prediction. - Implements two parallel CNNs plus a feed-forward fully connected network over protein and molecular sequence/structure representations. - Includes: - chemogenomic representative K-fold splitting; - Regression Discriminative Localization / L-Grad-RAM; - global max/average pooling and guided/non-guided gradient variants; - DAVIS kinase-affinity data and representations; - protein Smith–Waterman similarity; - SMILES ECFP6/Tanimoto similarity; - sc-PDB matching and PSSM data; - PyMOL docking sessions for ABL1(E255K)-phosphorylated/SKI-606 and DDR1/Foretinib; - training, validation, evaluation, and Docker instructions; - project-summary video and slide deck. - This appears inside `kganeshchandan/d4-course-projs` as well. Treat it as one course project with an organization repository and personal aggregate copy. --- ## 5. D4-course/Symmetry-aware-actor-critic - URL: . - Eight commits, all attributed to Ganesh. - Contribution period: 2022-11-30 through 2022-12-10. - Course import/adaptation of **MolGym: Reinforcement Learning for 3D Molecular Design**. - The upstream method trains RL policies that place atoms in Cartesian coordinates, with single-bag, stochastic-bag, and multi-bag experiments. - Repository adds course-oriented Docker instructions, a summary video, and a slide deck. - The README explicitly credits the original MolGym and Symmetry-Aware Actor-Critic authors. Portfolio wording must describe this as reproduction/adaptation/course work—not an original invention of the underlying algorithm. - This appears as the `molgym` directory inside `kganeshchandan/d4-course-projs`. --- ## 6. jai-ganesh-sankar/ML4NS - URL: . - Collaborative Machine Learning for Natural Sciences course-progress repository, Spring 2022. - Repository history: 12 commits; eight attributed to Ganesh. - Ganesh contribution period: 2022-04-13 through 2022-05-18. - Commit and tree evidence includes: - drug–drug and chemical-similarity datasets; - drug target, enzyme, pathway, transporter, indication, side-effect, and off-side-effect matrices; - graph embedding, Node2Vec, and t-SNE work; - CNN, residual CNN, CRNN, and feed-forward notebooks; - saved models; - a committed filename/report indicating CNN accuracy around 0.888 and a Ganesh commit message stating best CNN/FNN accuracy of 0.89. - Because the README supplies little task/evaluation context, keep this in the archival/course timeline unless the exact prediction task and metric are confirmed. --- ## 7. Virtual Labs experiments All three repositories explicitly list **Ganesh Chandan** and **Jai Ganesh** as IIIT-H developers, with Dr. Prabhakar Bhimalapuram as subject-matter expert. The repositories include theory, aims, procedures, pre/post-tests, Three.js simulations, and deployment workflows. ### virtual-labs/exp-defects-in-crystals-iiith - URL: . - Legacy slug says “defects in crystals,” but current experiment content identifies it as **Reciprocal Lattices**. - Ganesh commits: four, from 2023-01-09 through 2023-03-07. - Contributions include initializing the experiment, bug fixes, an observation section, instructions, and canvas/toggle/slider updates. - Experiment teaches primitive/reciprocal vectors for finite visualizations of simple cubic, body-centered cubic, face-centered cubic, and hexagonal close-packed lattices. ### virtual-labs/exp-miller-indices-iiith - URL: . - Experiment: **Miller Indices**. - Ganesh commits: four, from 2023-01-09 through 2023-03-07. - Contributions include Miller-index implementation/naming, bug fixes, observation section, and canvas/toggle/slider updates. - The simulation lets students select atoms, construct a plane, and check whether it matches selected Miller indices. ### virtual-labs/exp-symmetries-iiith - URL: . - Experiment: **Symmetries in crystal structures**. - Ganesh commits: 16, from 2023-01-11 through 2023-11-12. - Contributions include reference atoms, rotational controls, template testing, action handling, observation content, point/axis/plane symmetry checks, visualization, UI changes, axis rendering, overflow fixes, and ghost-rotation fixes. - Covers simple cubic, face-centered cubic, body-centered cubic, and hexagonal close-packed lattices. The repository histories continue into 2026 through other contributors/maintenance; that later activity is not attributed to Ganesh. --- ## Portfolio interpretation ### Flagship evidence - `devalab/MolGPT2.0` - `devalab/Protein-Ligand-Dataset-Bias` - `devalab/Spectra2Structure` - The three Virtual Labs experiments ### Course/archive evidence - `D4-course/EDDTR` - `D4-course/Symmetry-aware-actor-critic` - `jai-ganesh-sankar/ML4NS` ### Duplicate/mirror relationships - `devalab/Spectra2Structure` and `kganeshchandan/Spectra2Structure` are copies/variants of one research implementation. - `devalab/Protein-Ligand-Dataset-Bias`, `kganeshchandan/dd_code`, and `kganeshchandan/sim-CNNDTA` belong to the same binding-affinity bias research line but contain differing experiment coverage. - `D4-course/EDDTR` and `D4-course/Symmetry-aware-actor-critic` also appear inside the aggregate `kganeshchandan/d4-course-projs` repository. - `kganeshchandan/3JS` is a prototype/working repository related to the later organization-hosted Virtual Labs experiments.